MIX-MHingeGAN-CIFAR-10.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "[GPU] MIX-MHingeGAN-CIFAR-10.ipynb",
      "provenance": [],
      "collapsed_sections": [],
      "machine_shape": "hm",
      "mount_file_id": "1RbpVxU6sF2qhDD0EV1yP_WjCeIf_tBKq",
      "authorship_tag": "ABX9TyP/IsDrXy/3JSOvCj7Fk7dJ",
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/tsc2017/MIX-GAN/blob/main/MIX-MHingeGAN-CIFAR-10.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "i2KyBv_2qxmK",
        "outputId": "778953e1-6496-4f8d-b268-46100f8a8764",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 372
        }
      },
      "source": [
        "!nvidia-smi"
      ],
      "execution_count": 1,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Mon Oct 19 03:06:39 2020       \n",
            "+-----------------------------------------------------------------------------+\n",
            "| NVIDIA-SMI 455.23.05    Driver Version: 418.67       CUDA Version: 10.1     |\n",
            "|-------------------------------+----------------------+----------------------+\n",
            "| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |\n",
            "| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |\n",
            "|                               |                      |               MIG M. |\n",
            "|===============================+======================+======================|\n",
            "|   0  Tesla V100-SXM2...  Off  | 00000000:00:04.0 Off |                    0 |\n",
            "| N/A   34C    P0    25W / 300W |      0MiB / 16130MiB |      0%      Default |\n",
            "|                               |                      |                 ERR! |\n",
            "+-------------------------------+----------------------+----------------------+\n",
            "                                                                               \n",
            "+-----------------------------------------------------------------------------+\n",
            "| Processes:                                                                  |\n",
            "|  GPU   GI   CI        PID   Type   Process name                  GPU Memory |\n",
            "|        ID   ID                                                   Usage      |\n",
            "|=============================================================================|\n",
            "|  No running processes found                                                 |\n",
            "+-----------------------------------------------------------------------------+\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "VVMsjRIZrbO2",
        "outputId": "138b8a0a-2228-4d18-b870-30382557ae07",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        }
      },
      "source": [
        "# Mounting Google Drive ↖here would be simpler if you are the creator of a notebook. Maybe you can create a new notebook and copy the cells.\n",
        "from google.colab import drive\n",
        "drive.mount('/content/drive/')"
      ],
      "execution_count": 2,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Drive already mounted at /content/drive/; to attempt to forcibly remount, call drive.mount(\"/content/drive/\", force_remount=True).\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "TB_9wmqKJ4Sr"
      },
      "source": [
        "# Monitor GPU Utilization, GPU temp, etc\n",
        "#!pip install wandb\n",
        "#import wandb\n",
        "#wandb.init()"
      ],
      "execution_count": 3,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "u5VOpNC-Jlr4",
        "outputId": "4c34d4a9-34a8-4788-d61f-ae2edcaf299b",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 283
        }
      },
      "source": [
        "pip install tensorflow-gan"
      ],
      "execution_count": 4,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Collecting tensorflow-gan\n",
            "\u001b[?25l  Downloading https://files.pythonhosted.org/packages/0c/2e/62922111d7d50e1900e3030764743ea7735540ce103b3ab30fd5cd2d8a2b/tensorflow_gan-2.0.0-py2.py3-none-any.whl (365kB)\n",
            "\r\u001b[K     |█                               | 10kB 24.5MB/s eta 0:00:01\r\u001b[K     |█▉                              | 20kB 6.3MB/s eta 0:00:01\r\u001b[K     |██▊                             | 30kB 7.3MB/s eta 0:00:01\r\u001b[K     |███▋                            | 40kB 8.1MB/s eta 0:00:01\r\u001b[K     |████▌                           | 51kB 6.2MB/s eta 0:00:01\r\u001b[K     |█████▍                          | 61kB 5.9MB/s eta 0:00:01\r\u001b[K     |██████▎                         | 71kB 5.9MB/s eta 0:00:01\r\u001b[K     |███████▏                        | 81kB 6.2MB/s eta 0:00:01\r\u001b[K     |████████                        | 92kB 6.0MB/s eta 0:00:01\r\u001b[K     |█████████                       | 102kB 6.0MB/s eta 0:00:01\r\u001b[K     |█████████▉                      | 112kB 6.0MB/s eta 0:00:01\r\u001b[K     |██████████▊                     | 122kB 6.0MB/s eta 0:00:01\r\u001b[K     |███████████▋                    | 133kB 6.0MB/s eta 0:00:01\r\u001b[K     |████████████▋                   | 143kB 6.0MB/s eta 0:00:01\r\u001b[K     |█████████████▌                  | 153kB 6.0MB/s eta 0:00:01\r\u001b[K     |██████████████▍                 | 163kB 6.0MB/s eta 0:00:01\r\u001b[K     |███████████████▎                | 174kB 6.0MB/s eta 0:00:01\r\u001b[K     |████████████████▏               | 184kB 6.0MB/s eta 0:00:01\r\u001b[K     |█████████████████               | 194kB 6.0MB/s eta 0:00:01\r\u001b[K     |██████████████████              | 204kB 6.0MB/s eta 0:00:01\r\u001b[K     |██████████████████▉             | 215kB 6.0MB/s eta 0:00:01\r\u001b[K     |███████████████████▊            | 225kB 6.0MB/s eta 0:00:01\r\u001b[K     |████████████████████▋           | 235kB 6.0MB/s eta 0:00:01\r\u001b[K     |█████████████████████▌          | 245kB 6.0MB/s eta 0:00:01\r\u001b[K     |██████████████████████▍         | 256kB 6.0MB/s eta 0:00:01\r\u001b[K     |███████████████████████▎        | 266kB 6.0MB/s eta 0:00:01\r\u001b[K     |████████████████████████▎       | 276kB 6.0MB/s eta 0:00:01\r\u001b[K     |█████████████████████████▏      | 286kB 6.0MB/s eta 0:00:01\r\u001b[K     |██████████████████████████      | 296kB 6.0MB/s eta 0:00:01\r\u001b[K     |███████████████████████████     | 307kB 6.0MB/s eta 0:00:01\r\u001b[K     |███████████████████████████▉    | 317kB 6.0MB/s eta 0:00:01\r\u001b[K     |████████████████████████████▊   | 327kB 6.0MB/s eta 0:00:01\r\u001b[K     |█████████████████████████████▋  | 337kB 6.0MB/s eta 0:00:01\r\u001b[K     |██████████████████████████████▌ | 348kB 6.0MB/s eta 0:00:01\r\u001b[K     |███████████████████████████████▍| 358kB 6.0MB/s eta 0:00:01\r\u001b[K     |████████████████████████████████| 368kB 6.0MB/s \n",
            "\u001b[?25hRequirement already satisfied: tensorflow-hub>=0.2 in /usr/local/lib/python3.6/dist-packages (from tensorflow-gan) (0.9.0)\n",
            "Requirement already satisfied: tensorflow-probability>=0.7 in /usr/local/lib/python3.6/dist-packages (from tensorflow-gan) (0.11.0)\n",
            "Requirement already satisfied: numpy>=1.12.0 in /usr/local/lib/python3.6/dist-packages (from tensorflow-hub>=0.2->tensorflow-gan) (1.18.5)\n",
            "Requirement already satisfied: six>=1.12.0 in /usr/local/lib/python3.6/dist-packages (from tensorflow-hub>=0.2->tensorflow-gan) (1.15.0)\n",
            "Requirement already satisfied: protobuf>=3.8.0 in /usr/local/lib/python3.6/dist-packages (from tensorflow-hub>=0.2->tensorflow-gan) (3.12.4)\n",
            "Requirement already satisfied: gast>=0.3.2 in /usr/local/lib/python3.6/dist-packages (from tensorflow-probability>=0.7->tensorflow-gan) (0.3.3)\n",
            "Requirement already satisfied: decorator in /usr/local/lib/python3.6/dist-packages (from tensorflow-probability>=0.7->tensorflow-gan) (4.4.2)\n",
            "Requirement already satisfied: dm-tree in /usr/local/lib/python3.6/dist-packages (from tensorflow-probability>=0.7->tensorflow-gan) (0.1.5)\n",
            "Requirement already satisfied: cloudpickle==1.3 in /usr/local/lib/python3.6/dist-packages (from tensorflow-probability>=0.7->tensorflow-gan) (1.3.0)\n",
            "Requirement already satisfied: setuptools in /usr/local/lib/python3.6/dist-packages (from protobuf>=3.8.0->tensorflow-hub>=0.2->tensorflow-gan) (50.3.0)\n",
            "Installing collected packages: tensorflow-gan\n",
            "Successfully installed tensorflow-gan-2.0.0\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "b3BbtNdwQuw7"
      },
      "source": [
        "# GPUs with Tensor Cores can benefit from mixed-precision training substancially\n",
        "!export TF_ENABLE_AUTO_MIXED_PRECISION=1"
      ],
      "execution_count": 5,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "iqgBrY57JmUh"
      },
      "source": [
        "import os, sys\n",
        "# The \"tflib\" directory is in os.environ['HOME']; the datasets are in os.environ['DATA']. The log directory will be created automatically if it does not exist.\n",
        "os.environ['HOME']=\"/content/drive/Shared drives/shared/mix-gan\" # My Drive or Shared drives\n",
        "os.environ['DATA']=\"/content/drive/Shared drives/shared/mix-gan/data\" \n",
        "os.environ['LOG']=\"/content/drive/Shared drives/shared/mix-gan/logs\" "
      ],
      "execution_count": 6,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "qNfTqNF8KEFh",
        "outputId": "a0260f4f-162f-4592-e0b4-43f2f2971fb2",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 52
        }
      },
      "source": [
        "%cd /content/drive/Shared drives/shared/mix-gan/data\n",
        "# Download and extract the CIFAR-10 dataset automatically if it does not exist\n",
        "if not os.path.isfile(os.environ['DATA']+'/cifar-10-python.tar.gz'):\n",
        "  !wget www.cs.toronto.edu/~kriz/cifar-10-python.tar.gz\n",
        "  !tar vxzf cifar-10-python.tar.gz\n",
        "%cd .."
      ],
      "execution_count": 7,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "/content/drive/Shared drives/shared/mix-gan/data\n",
            "/content/drive/Shared drives/shared/mix-gan\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "YZDHUw-pJQQP",
        "outputId": "93d15c16-6f5b-4dae-aae3-d485c0ab1038",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        }
      },
      "source": [
        "import tensorflow.compat.v1 as tf\n",
        "import os, sys\n",
        "sys.path.append(os.getcwd())\n",
        "gpu_id=[0]\n",
        "DEVICES = ['/gpu:{}'.format(i) for i in range(len(gpu_id))]\n",
        "os.environ[\"CUDA_DEVICE_ORDER\"]=\"PCI_BUS_ID\"   # see issue #152\n",
        "os.environ[\"CUDA_VISIBLE_DEVICES\"]=','.join([str(i) for i in gpu_id]) #set gpu id\n",
        "import random\n",
        "import time\n",
        "\n",
        "import functools\n",
        "import matplotlib\n",
        "matplotlib.use('Agg')\n",
        "import matplotlib.pyplot as plt\n",
        "import numpy as np\n",
        "import sklearn.datasets\n",
        "import scipy.misc\n",
        "import urllib\n",
        "import tflib as lib\n",
        "import tflib.ops.linear\n",
        "import tflib.ops.conv2d\n",
        "import tflib.ops.batchnorm\n",
        "import tflib.ops.cond_batchnorm\n",
        "import tflib.ops.layernorm\n",
        "import tflib.ops.deconv2d\n",
        "import tflib.save_images\n",
        "import tflib.mnist\n",
        "import tflib.cifar10\n",
        "#import tflib.cifar100\n",
        "import tflib.imagenet64\n",
        "import tflib.imagenet128_10classes\n",
        "import tflib.imagenet128\n",
        "import tflib.small_imagenet\n",
        "#import tflib.stl\n",
        "from tflib.ada import augment\n",
        "import tflib.plot\n",
        "import tflib.inception_score\n",
        "import tflib.fid\n",
        "import errno\n",
        "#from cleverhans.attacks_tf import fgsm\n",
        "import locale\n",
        "dtype='float32'\n",
        "\n",
        "DATASET = 'CIFAR-10' #CIFAR, small_ImageNet, ImageNet64, ImageNet128_10classes, ImageNet128, STL\n",
        "\n",
        "gan_learning_rate=2e-4\n",
        "\n",
        "NUM_GENERATORS=10\n",
        "NUM_DISCRIMINATORS=10\n",
        "\n",
        "# For WGAN-GP\n",
        "BASE_DIM=8\n",
        "G_DIM = 32*BASE_DIM # Generator dimensionality\n",
        "D_DIM = 32*BASE_DIM  # Discriminator dimensionality\n",
        "\n",
        "RAND_LABELS=False\n",
        "SHARED_EMBED=1\n",
        "PL_COEFF=0\n",
        "HIER=1\n",
        "ATTENTION=0\n",
        "NUM_RESIDUAL_UNIT=1\n",
        "BOTTLENECK=0\n",
        "GET_LABELS=True\n",
        "BATCH_SIZE = 50 # Batch size, must be divisible by the dataset size if using MNIST \n",
        "GET_INCEPTION_SCORE=1\n",
        "GET_FID=1\n",
        "INCEPTION_BATCHES=int(50000/BATCH_SIZE)\n",
        "FID_BATCHES=int(50000/BATCH_SIZE)\n",
        "TRAIN_SIZE=BATCH_SIZE*1 # Define the training set size, ignored if using a realistic image dataset\n",
        "TEST_SIZE=BATCH_SIZE*1 # Define the test set size, ignored if using a realistic image dataset\n",
        "LAMBDA =0 # Smaller lambda makes things faster for toy tasks, but isn't necessary if you increase N_CRITIC enough\n",
        "R1_COEFF=0\n",
        "################################################\n",
        "GAN_LOSS=False\n",
        "HINGE_LOSS=False\n",
        "WGAN_LOSS=False\n",
        "AM_LOSS=False\n",
        "MH_LOSS=True\n",
        "assert GAN_LOSS + HINGE_LOSS + WGAN_LOSS + AM_LOSS + MH_LOSS == 1\n",
        "############################################\n",
        "#settings\n",
        "ADA=False\n",
        "ADA_TARGET=.6\n",
        "REC_NORM=2 # useless and need not be changed\n",
        "REC_COEFF=0\n",
        "ORTH_COEFF=0\n",
        "LAMBDA3=0.05\n",
        "Z_DIM=120\n",
        "\n",
        "N_GENERATOR = 1 # Generator steps per general steps\n",
        "N_CRITIC = 4 # Critic steps per general steps\n",
        "ITERS=100000\n",
        "EMA=True\n",
        "EMA_ITER=2000\n",
        "ACCU_STATS=True\n",
        "snapshot_interval=5000\n",
        "log_interval=1000\n",
        "num_modal=3\n",
        "std=.0# noise strength in CMNIST\n",
        "\n",
        "\n",
        "LAMBDA2 = 0  # parameter LAMBDA2 for CT-GAN\n",
        "DROPOUT=1 if LAMBDA2>0 else 0\n",
        "Factor_M = 0.0  # factor M\n",
        "NORMALIZATION_G = True # Use batchnorm in generator?\n",
        "NORMALIZATION_D = False # Use batchnorm (or layernorm) in critic?\n",
        "DECAY = True # Whether to decay LR over learning\n",
        "SPECTRAL_NORMALIZATION = True\n",
        "CONDITIONAL = True  # Whether to train a conditional or unconditional model\n",
        "\n",
        "ACGAN_SCALE = 0 # How to scale the critic's ACGAN loss relative to WGAN loss\n",
        "ACGAN_SCALE_G = .0 # How to scale generator's ACGAN loss relative to WGAN loss\n",
        "ACGAN = ACGAN_SCALE!=0\n",
        "\n",
        "BETA1=0.\n",
        "BETA2=.9\n",
        "STYLEGAN=False\n",
        "STYLEGAN2=False\n",
        "if STYLEGAN or STYLEGAN2: Z_DIM=512\n",
        "if STYLEGAN2:\n",
        "    GAN_LOSS=1\n",
        "    MH_LOSS=False\n",
        "    HINGE_LOSS=False\n",
        "    WGAN_LOSS=0\n",
        "    AM_LOSS=False\n",
        "    R1_COEFF=0.01\n",
        "    RAND_LABELS=True\n",
        "    ACCU_STATS=False\n",
        "    LAMBDA3=0\n",
        "    N_CRITIC = 1\n",
        "    BATCH_SIZE=64\n",
        "    BETA2= .99\n",
        "    gan_learning_rate=2.5e-3\n",
        "    ITERS=100e6//BATCH_SIZE\n",
        "COMPUTE_GRADIENT = LAMBDA>0 or R1_COEFF>0\n",
        "locale.setlocale(locale.LC_ALL, '')\n",
        "lib.print_model_settings(locals().copy())"
      ],
      "execution_count": 8,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "WARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/tensorflow/python/compat/v2_compat.py:96: disable_resource_variables (from tensorflow.python.ops.variable_scope) is deprecated and will be removed in a future version.\n",
            "Instructions for updating:\n",
            "non-resource variables are not supported in the long term\n",
            "WARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/tensorflow_gan/python/estimator/tpu_gan_estimator.py:42: The name tf.estimator.tpu.TPUEstimator is deprecated. Please use tf.compat.v1.estimator.tpu.TPUEstimator instead.\n",
            "\n",
            "Device mapping:\n",
            "/job:localhost/replica:0/task:0/device:XLA_CPU:0 -> device: XLA_CPU device\n",
            "/job:localhost/replica:0/task:0/device:XLA_GPU:0 -> device: XLA_GPU device\n",
            "/job:localhost/replica:0/task:0/device:GPU:0 -> device: 0, name: Tesla V100-SXM2-16GB, pci bus id: 0000:00:04.0, compute capability: 7.0\n",
            "\n",
            "WARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/tensorflow/python/util/nest.py:635: flatten (from tensorflow.python.keras.legacy_tf_layers.core) is deprecated and will be removed in a future version.\n",
            "Instructions for updating:\n",
            "Use keras.layers.Flatten instead.\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "WARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/tensorflow/python/util/nest.py:635: flatten (from tensorflow.python.keras.legacy_tf_layers.core) is deprecated and will be removed in a future version.\n",
            "Instructions for updating:\n",
            "Use keras.layers.Flatten instead.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "WARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/tensorflow/python/keras/legacy_tf_layers/core.py:332: Layer.apply (from tensorflow.python.keras.engine.base_layer_v1) is deprecated and will be removed in a future version.\n",
            "Instructions for updating:\n",
            "Please use `layer.__call__` method instead.\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "WARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/tensorflow/python/keras/legacy_tf_layers/core.py:332: Layer.apply (from tensorflow.python.keras.engine.base_layer_v1) is deprecated and will be removed in a future version.\n",
            "Instructions for updating:\n",
            "Please use `layer.__call__` method instead.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "Device mapping:\n",
            "/job:localhost/replica:0/task:0/device:XLA_CPU:0 -> device: XLA_CPU device\n",
            "/job:localhost/replica:0/task:0/device:XLA_GPU:0 -> device: XLA_GPU device\n",
            "/job:localhost/replica:0/task:0/device:GPU:0 -> device: 0, name: Tesla V100-SXM2-16GB, pci bus id: 0000:00:04.0, compute capability: 7.0\n",
            "\n",
            "Uppercase local vars:\n",
            "\tACCU_STATS: True\n",
            "\tACGAN: False\n",
            "\tACGAN_SCALE: 0\n",
            "\tACGAN_SCALE_G: 0.0\n",
            "\tADA: False\n",
            "\tADA_TARGET: 0.6\n",
            "\tAM_LOSS: False\n",
            "\tATTENTION: 0\n",
            "\tBASE_DIM: 8\n",
            "\tBATCH_SIZE: 50\n",
            "\tBETA1: 0.0\n",
            "\tBETA2: 0.9\n",
            "\tBOTTLENECK: 0\n",
            "\tCOMPUTE_GRADIENT: False\n",
            "\tCONDITIONAL: True\n",
            "\tDATASET: CIFAR-10\n",
            "\tDECAY: True\n",
            "\tDEVICES: ['/gpu:0']\n",
            "\tDROPOUT: 0\n",
            "\tD_DIM: 256\n",
            "\tEMA: True\n",
            "\tEMA_ITER: 2000\n",
            "\tFID_BATCHES: 1000\n",
            "\tGAN_LOSS: False\n",
            "\tGET_FID: 1\n",
            "\tGET_INCEPTION_SCORE: 1\n",
            "\tGET_LABELS: True\n",
            "\tG_DIM: 256\n",
            "\tHIER: 1\n",
            "\tHINGE_LOSS: False\n",
            "\tINCEPTION_BATCHES: 1000\n",
            "\tITERS: 100000\n",
            "\tLAMBDA: 0\n",
            "\tLAMBDA2: 0\n",
            "\tLAMBDA3: 0.05\n",
            "\tMH_LOSS: True\n",
            "\tNORMALIZATION_D: False\n",
            "\tNORMALIZATION_G: True\n",
            "\tNUM_DISCRIMINATORS: 10\n",
            "\tNUM_GENERATORS: 10\n",
            "\tNUM_RESIDUAL_UNIT: 1\n",
            "\tN_CRITIC: 4\n",
            "\tN_GENERATOR: 1\n",
            "\tORTH_COEFF: 0\n",
            "\tPL_COEFF: 0\n",
            "\tR1_COEFF: 0\n",
            "\tRAND_LABELS: False\n",
            "\tREC_COEFF: 0\n",
            "\tREC_NORM: 2\n",
            "\tSHARED_EMBED: 1\n",
            "\tSPECTRAL_NORMALIZATION: True\n",
            "\tSTYLEGAN: False\n",
            "\tSTYLEGAN2: False\n",
            "\tTEST_SIZE: 50\n",
            "\tTRAIN_SIZE: 50\n",
            "\tWGAN_LOSS: False\n",
            "\tZ_DIM: 120\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "9KtwmlDJJQQV",
        "outputId": "d36e1727-1661-4f9d-9c6d-7b3fddea4e99",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        }
      },
      "source": [
        "tf.__version__"
      ],
      "execution_count": 9,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "application/vnd.google.colaboratory.intrinsic+json": {
              "type": "string"
            },
            "text/plain": [
              "'2.3.0'"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 9
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "axLYJLaEJQQZ"
      },
      "source": [
        ""
      ],
      "execution_count": 9,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "P6ggh59AJQQd"
      },
      "source": [
        "#Datasets\n",
        "IMAGE_DATASETS=['MNIST','CMNIST','CIFAR-10', 'ImageNet64', 'ImageNet128_10classes', 'ImageNet128','small_ImageNet','STL','STL32','STL48','STL64','STL128','Zap50k']\n",
        "ENCODER_MODELS=['ALI','IDGAN']\n",
        "\n",
        "assert DATASET in IMAGE_DATASETS or NETWORK_TYPE is 'FCGAN', 'Cannot use DCGAN for non-image datasets.'\n",
        "\n",
        "if DATASET=='MNIST':\n",
        "    NUM_CHANNELS=1\n",
        "    HEIGHT=WIDTH=32\n",
        "    DATA_DIM=HEIGHT*WIDTH*NUM_CHANNELS\n",
        "    TRAIN_SIZE=50000\n",
        "    TEST_SIZE=10000\n",
        "if DATASET=='CMNIST':\n",
        "    NUM_CHANNELS=3\n",
        "    HEIGHT=WIDTH=32\n",
        "    DATA_DIM=HEIGHT*WIDTH*NUM_CHANNELS\n",
        "    SKETCH_DIM=HEIGHT*WIDTH\n",
        "    TRAIN_SIZE=50000\n",
        "    TEST_SIZE=10000\n",
        "if DATASET=='CIFAR-10':\n",
        "    NUM_CHANNELS=3\n",
        "    HEIGHT=WIDTH=32\n",
        "    DATA_DIM=HEIGHT*WIDTH*NUM_CHANNELS\n",
        "    NUM_CLASSES=10\n",
        "    CODE_DIM=10\n",
        "    TRAIN_SIZE=50000\n",
        "    TEST_SIZE=10000\n",
        "    # Download CIFAR-10 (Python version) at\n",
        "    # https://www.cs.toronto.edu/~kriz/cifar.html and fill in the path to the\n",
        "    # extracted files here!\n",
        "    DATA_DIR = os.environ['DATA']+'/cifar-10-batches-py/'#directory to save the dataset\n",
        "\n",
        "if DATASET=='ImageNet64':\n",
        "    NUM_CHANNELS=3\n",
        "    HEIGHT=WIDTH=64\n",
        "    DATA_DIM=HEIGHT*WIDTH*NUM_CHANNELS\n",
        "    NUM_CLASSES=1000\n",
        "    CODE_DIM=NUM_CLASSES\n",
        "    TRAIN_SIZE=1281167\n",
        "    TEST_SIZE=50000\n",
        "    DATA_DIR = os.environ['DATA']+'/ImageNet64/'#directory to save the dataset\n",
        "    \n",
        "if DATASET=='ImageNet128_10classes':\n",
        "    NUM_CHANNELS=3\n",
        "    HEIGHT=WIDTH=128\n",
        "    DATA_DIM=HEIGHT*WIDTH*NUM_CHANNELS\n",
        "    NUM_CLASSES=10\n",
        "    CODE_DIM=NUM_CLASSES\n",
        "    TRAIN_SIZE=12697\n",
        "    TEST_SIZE=500\n",
        "    DATA_DIR = os.environ['DATA']+'/ImageNet128_10classes/'#directory to save the dataset\n",
        "    \n",
        "if DATASET=='ImageNet128':\n",
        "    NUM_CHANNELS=3\n",
        "    HEIGHT=WIDTH=128\n",
        "    DATA_DIM=HEIGHT*WIDTH*NUM_CHANNELS\n",
        "    NUM_CLASSES=1000\n",
        "    CODE_DIM=NUM_CLASSES\n",
        "    TRAIN_SIZE=1281159\n",
        "    TEST_SIZE=50000\n",
        "    DATA_DIR = os.environ['DATA']+'/ImageNet128/'#directory to save the dataset\n",
        "    \n",
        "if DATASET=='small_ImageNet':\n",
        "    NUM_CHANNELS=3\n",
        "    HEIGHT=WIDTH=64\n",
        "    DATA_DIM=HEIGHT*WIDTH*NUM_CHANNELS\n",
        "    NUM_CLASSES=1000\n",
        "    CODE_DIM=NUM_CLASSES\n",
        "    TRAIN_SIZE=1281150\n",
        "    TEST_SIZE=50000\n",
        "    # Download 64x64 small_ImageNet at http://image-net.org/small/download.php and\n",
        "    # fill in the path to the extracted files here!\n",
        "    DATA_DIR = os.environ['DATA']+'/small_ImageNet/' #directory to save the dataset\n",
        "\n",
        "if DATASET=='STL':\n",
        "    NUM_CHANNELS=3\n",
        "    HEIGHT=WIDTH=96\n",
        "    DATA_DIM=HEIGHT*WIDTH*NUM_CHANNELS\n",
        "    NUM_CLASSES=10\n",
        "    CODE_DIM=10\n",
        "    TRAIN_SIZE=5000\n",
        "    TEST_SIZE=8000\n",
        "    # fill in the path to the extracted files here!\n",
        "    DATA_DIR = os.environ['DATA']+'/STL-10/stl10_matlab/' #directory to save the dataset\n",
        "    \n",
        "if DATASET=='STL32':\n",
        "    NUM_CHANNELS=3\n",
        "    HEIGHT=WIDTH=32\n",
        "    DATA_DIM=HEIGHT*WIDTH*NUM_CHANNELS\n",
        "    NUM_CLASSES=10\n",
        "    CODE_DIM=10\n",
        "    TRAIN_SIZE=5000\n",
        "    TEST_SIZE=8000\n",
        "    # fill in the path to the extracted files here!\n",
        "    DATA_DIR = os.environ['DATA']+'/STL-10/stl10_matlab/' #directory to save the dataset    \n",
        "if DATASET=='STL48':\n",
        "    NUM_CHANNELS=3\n",
        "    HEIGHT=WIDTH=48\n",
        "    DATA_DIM=HEIGHT*WIDTH*NUM_CHANNELS\n",
        "    NUM_CLASSES=10\n",
        "    CODE_DIM=10\n",
        "    TRAIN_SIZE=5000\n",
        "    TEST_SIZE=8000\n",
        "    # fill in the path to the extracted files here!\n",
        "    DATA_DIR = os.environ['DATA']+'/STL-10/stl10_matlab/' #directory to save the dataset  \n",
        "if DATASET=='STL64':\n",
        "    NUM_CHANNELS=3\n",
        "    HEIGHT=WIDTH=64\n",
        "    DATA_DIM=HEIGHT*WIDTH*NUM_CHANNELS\n",
        "    NUM_CLASSES=10\n",
        "    CODE_DIM=10\n",
        "    TRAIN_SIZE=5000\n",
        "    TEST_SIZE=8000\n",
        "    # fill in the path to the extracted files here!\n",
        "    DATA_DIR = os.environ['DATA']+'/STL-10/stl10_matlab/' #directory to save the dataset  \n",
        "if DATASET=='STL128':\n",
        "    NUM_CHANNELS=3\n",
        "    HEIGHT=WIDTH=128\n",
        "    DATA_DIM=HEIGHT*WIDTH*NUM_CHANNELS\n",
        "    NUM_CLASSES=10\n",
        "    CODE_DIM=10\n",
        "    TRAIN_SIZE=5000\n",
        "    TEST_SIZE=8000\n",
        "    # fill in the path to the extracted files here!\n",
        "    DATA_DIR = os.environ['DATA']+'/STL-10/stl10_matlab/' #directory to save the dataset  \n",
        "if DATASET=='Zap50k':\n",
        "    NUM_CHANNELS=3\n",
        "    REAL_HEIGHT=102\n",
        "    REAL_WIDTH=136\n",
        "    HEIGHT=WIDTH=128\n",
        "    DATA_DIM=HEIGHT*WIDTH*NUM_CHANNELS\n",
        "    SKETCH_DIM=HEIGHT*WIDTH\n",
        "    TRAIN_SIZE=40000\n",
        "    TEST_SIZE=10025\n",
        "    DATA_DIR = os.environ['DATA']+'/ut-zap50k/ut-zap50k-images-unfolded'#directory to save the dataset\n",
        "    SKETCH_DIR =os.environ['DATA']+'/ut-zap50k/ut-zap50k-sketches-unfolded'\n",
        "\n",
        "if DATASET in IMAGE_DATASETS:\n",
        "    G_LAYERS=int(np.ceil(np.log2(HEIGHT/4)))+2\n",
        "\n",
        "# Dataset iterator\n",
        "def inf_gen(MODE='TRAIN'):\n",
        "    if DATASET == '25gaussians':\n",
        "    \n",
        "        dataset = []\n",
        "        for i in range(100000/25):\n",
        "            for x in range(-2, 3):\n",
        "                for y in range(-2, 3):\n",
        "                    point = np.random.randn(2)*0.05\n",
        "                    point[0] += 2*x\n",
        "                    point[1] += 2*y\n",
        "                    dataset.append(point)\n",
        "        dataset = np.array(dataset, dtype=dtype)\n",
        "        np.random.shuffle(dataset)\n",
        "        dataset /= 2.828 # stdev\n",
        "        while True:\n",
        "            for i in range(len(dataset)/BATCH_SIZE):\n",
        "                yield dataset[i*BATCH_SIZE:(i+1)*BATCH_SIZE]\n",
        "\n",
        "    elif DATASET == 'swissroll':\n",
        "\n",
        "        while True:\n",
        "            data = sklearn.datasets.make_swiss_roll(\n",
        "                n_samples=BATCH_SIZE, \n",
        "                ran_noise=0.25\n",
        "            )[0]\n",
        "            data = data.astype(dtype)[:, [0, 2]]\n",
        "            data /= 7.5 # stdev plus a little\n",
        "            yield data\n",
        "\n",
        "    elif DATASET == '8gaussians':\n",
        "    \n",
        "        centers = [\n",
        "            (1,0),\n",
        "            (-1,0),\n",
        "            (0,1),\n",
        "            (0,-1),\n",
        "            (1./np.sqrt(2), 1./np.sqrt(2)),\n",
        "            (1./np.sqrt(2), -1./np.sqrt(2)),\n",
        "            (-1./np.sqrt(2), 1./np.sqrt(2)),\n",
        "            (-1./np.sqrt(2), -1./np.sqrt(2))\n",
        "        ]\n",
        "        centers = [(scale*x,scale*y) for x,y in centers]\n",
        "        while True:\n",
        "            dataset = []\n",
        "            for i in range(BATCH_SIZE):\n",
        "                point = np.random.randn(2)*.02\n",
        "                center = random.choice(centers)\n",
        "                point[0] += center[0]\n",
        "                point[1] += center[1]\n",
        "                dataset.append(point)\n",
        "            dataset = np.array(dataset, dtype=dtype)\n",
        "            dataset /= 1.414 # stdev\n",
        "            yield dataset\n",
        "\n",
        "            \n",
        "    elif DATASET == 'n-gaussians':         \n",
        "        centers = []\n",
        "        for i in range(num_modal):\n",
        "            centers.append(scale*tf.one_hot(i,DATA_DIM,dtype=dtype).eval())\n",
        "                    \n",
        "        if not FIX_DATA:\n",
        "            while True:\n",
        "                dataset = []\n",
        "                points = np.random.randn(BATCH_SIZE, DATA_DIM)*std\n",
        "                #points=tf.random_normal([BATCH_SIZE, DATA_DIM])*std\n",
        "                #center_choice=np.zeros((BATCH_SIZE, DATA_DIM), dtype='float32')\n",
        "                for i in range(BATCH_SIZE):\n",
        "                    points[i][random.randint(0,num_modal-1)]+=1.\n",
        "                #dataset = np.array(points, dtype='float32')\n",
        "                #dataset /= 1.414 # stdev\n",
        "                yield points\n",
        "        else:            \n",
        "            if MODE=='TRAIN':\n",
        "                fixed_train_data=np.random.randn(TRAIN_SIZE, DATA_DIM)*std\n",
        "                for i in range(TRAIN_SIZE):\n",
        "                    fixed_train_data[i][random.randint(0,num_modal-1)]+=1.\n",
        "                while True:\n",
        "                    for i in range(fixed_train_data.shape[0]/BATCH_SIZE):\n",
        "                        yield fixed_train_data[i*BATCH_SIZE:(i+1)*BATCH_SIZE]\n",
        "                        \n",
        "            elif MODE=='TEST':\n",
        "                fixed_test_data=np.random.randn(TEST_SIZE, DATA_DIM)*std\n",
        "                for i in range(TEST_SIZE):\n",
        "                    fixed_test_data[i][random.randint(0,num_modal-1)]+=1.\n",
        "                while True:\n",
        "                    for i in range(fixed_test_data.shape[0]/BATCH_SIZE):\n",
        "                        yield fixed_test_data[i*BATCH_SIZE:(i+1)*BATCH_SIZE]\n",
        "            else:\n",
        "                raise UnsupportedDataMode\n",
        "            \n",
        "    elif DATASET == 'ball':           \n",
        "        if not FIX_DATA:\n",
        "            while True:\n",
        "                points=np.random.randn(BATCH_SIZE,DATA_DIM)\n",
        "                dataset=points/np.linalg.norm(points,axis=1,keepdims=True)*scale\n",
        "                yield dataset\n",
        "        else:              \n",
        "            if MODE=='TRAIN':\n",
        "                fixed_train_data = np.random.randn(TRAIN_SIZE,DATA_DIM)\n",
        "                fixed_train_data=fixed_train_data/np.linalg.norm(fixed_train_data,axis=1,keepdims=True)*scale\n",
        "                while True:\n",
        "                    for i in range(fixed_train_data.shape[0]/BATCH_SIZE):\n",
        "                        yield fixed_train_data[i*BATCH_SIZE:(i+1)*BATCH_SIZE]\n",
        "            elif MODE=='TEST':\n",
        "                fixed_test_data = np.random.randn(TEST_SIZE,DATA_DIM)\n",
        "                fixed_test_data=fixed_test_data/np.linalg.norm(fixed_test_data,axis=1,keepdims=True)*scale\n",
        "                while True:\n",
        "                    for i in range(fixed_test_data.shape[0]/BATCH_SIZE):\n",
        "                        yield fixed_test_data[i*BATCH_SIZE:(i+1)*BATCH_SIZE]\n",
        "            else:\n",
        "                raise UnsupportedDataMode\n",
        "    elif DATASET == 'net': \n",
        "        if not FIX_DATA:\n",
        "            while True:\n",
        "                yield session.run(real_gen_data)\n",
        "        else:              \n",
        "            if MODE=='TRAIN':\n",
        "                fixed_train_data = Real_Gen(TRAIN_SIZE, real_data).eval()\n",
        "                while True:\n",
        "                    for i in range(fixed_train_data.shape[0]/BATCH_SIZE):\n",
        "                        yield fixed_train_data[i*BATCH_SIZE:(i+1)*BATCH_SIZE]\n",
        "            elif MODE=='TEST':\n",
        "                fixed_test_data = Real_Gen(TEST_SIZE, real_data).eval()\n",
        "                while True:\n",
        "                    for i in range(fixed_test_data.shape[0]/BATCH_SIZE):\n",
        "                        yield fixed_test_data[i*BATCH_SIZE:(i+1)*BATCH_SIZE]\n",
        "            else:\n",
        "                raise UnsupportedDataMode\n",
        "    elif DATASET=='MNIST':\n",
        "        if MODE=='TRAIN':\n",
        "            train_gen, _, _ = lib.mnist.load(BATCH_SIZE, BATCH_SIZE)\n",
        "            while True:\n",
        "                for images,targets in train_gen():\n",
        "                    yield images*2-1\n",
        "        elif MODE=='TEST':\n",
        "            _, val_gen, _ = lib.mnist.load(BATCH_SIZE, BATCH_SIZE)\n",
        "            while True:\n",
        "                for images,targets in val_gen():\n",
        "                    yield images*2-1\n",
        "    elif DATASET=='CMNIST':\n",
        "        if MODE=='TRAIN':\n",
        "            train_gen, _, _ = lib.mnist.load(BATCH_SIZE, BATCH_SIZE)\n",
        "            while True:\n",
        "                for images,targets in train_gen():\n",
        "                    if DENOISE:\n",
        "                        images=np.tile(images.reshape(-1,1,HEIGHT,WIDTH),(1,3,1,1))\n",
        "                        images=(images*np.random.rand(images.shape[0],3,1,1)).reshape(-1,DATA_DIM)*2-1\n",
        "                        images+=std*np.random.randn(images.shape[0],images.shape[1])\n",
        "                        images[images>1]=1\n",
        "                        images[images<-1]=-1\n",
        "                        yield images\n",
        "                    else:\n",
        "                        images=np.tile(images.reshape(-1,1,HEIGHT,WIDTH),(1,3,1,1))\n",
        "                        yield (images*np.random.rand(images.shape[0],3,1,1)).reshape(-1,DATA_DIM)*2-1                   \n",
        "        elif MODE=='TEST':\n",
        "            _, val_gen, _ = lib.mnist.load(BATCH_SIZE, BATCH_SIZE)\n",
        "            while True:\n",
        "                for images,targets in val_gen():\n",
        "                    if DENOISE:\n",
        "                        images=np.tile(images.reshape(-1,1,HEIGHT,WIDTH),(1,3,1,1))\n",
        "                        images=(images*np.random.rand(images.shape[0],3,1,1)).reshape(-1,DATA_DIM)*2-1\n",
        "                        images+=std*np.random.randn(images.shape[0],images.shape[1])\n",
        "                        images[images>1]=1\n",
        "                        images[images<-1]=-1\n",
        "                        yield images\n",
        "                    else:\n",
        "                        images=np.tile(images.reshape(-1,1,HEIGHT,WIDTH),(1,3,1,1))\n",
        "                        yield (images*np.random.rand(images.shape[0],3,1,1)).reshape(-1,DATA_DIM)*2-1  \n",
        "    elif DATASET=='CIFAR-10':\n",
        "        if MODE=='TRAIN':\n",
        "            train_gen, _ = lib.cifar10.load(BATCH_SIZE, data_dir=DATA_DIR)\n",
        "            while True:\n",
        "                for original_images, labels in train_gen():\n",
        "                    yield 2./255*original_images-1,labels\n",
        "        elif MODE=='TEST':\n",
        "            _, test_gen = lib.cifar10.load(BATCH_SIZE, data_dir=DATA_DIR)\n",
        "            while True:\n",
        "                for original_images, labels in test_gen():\n",
        "                    yield 2./255*original_images-1,labels\n",
        "    elif DATASET=='ImageNet64':\n",
        "            gen = lib.imagenet64.load(MODE, BATCH_SIZE, data_dir=DATA_DIR)\n",
        "            while True:\n",
        "                for original_images, labels in gen():\n",
        "                    yield 2./255*original_images-1,labels\n",
        "    elif DATASET=='ImageNet128':\n",
        "            gen = lib.imagenet128.load(MODE, BATCH_SIZE, data_dir=DATA_DIR)\n",
        "            while True:\n",
        "                for original_images, labels in gen():\n",
        "                    yield 2./255*original_images-1,labels\n",
        "    elif DATASET=='ImageNet128_10classes':\n",
        "            gen = lib.imagenet128_10classes.load(MODE, BATCH_SIZE, data_dir=DATA_DIR)\n",
        "            while True:\n",
        "                for original_images, labels in gen():\n",
        "                    yield 2./255*original_images-1,labels\n",
        "    elif DATASET=='small_ImageNet':\n",
        "        if MODE=='TRAIN':\n",
        "            train_gen, _ = lib.small_imagenet.load(BATCH_SIZE, data_dir=DATA_DIR)\n",
        "            while True:\n",
        "                for images in train_gen():\n",
        "                    yield images[0].reshape((-1,DATA_DIM))*2./255-1, np.zeros([BATCH_SIZE])\n",
        "        elif MODE=='TEST':\n",
        "            _, test_gen = lib.small_imagenet.load(BATCH_SIZE, data_dir=DATA_DIR)\n",
        "            while True:\n",
        "                for images in test_gen():\n",
        "                    yield images[0].reshape((-1,DATA_DIM))*2./255-1, np.zeros([BATCH_SIZE])\n",
        "    elif 'STL' in DATASET:\n",
        "            gen = lib.stl.load(MODE, BATCH_SIZE, data_dir=DATA_DIR, size=HEIGHT)\n",
        "            while True:\n",
        "                for original_images, labels in gen():\n",
        "                    yield 2./255*original_images-1,labels\n",
        "    elif DATASET=='Zap50k':\n",
        "        perm_matrices=np.zeros((6,3,3))\n",
        "        perm_matrices[0]=np.array([[1,0,0],[0,1,0],[0,0,1]])\n",
        "        perm_matrices[1]=np.array([[1,0,0],[0,0,1],[0,1,0]])\n",
        "        perm_matrices[2]=np.array([[0,1,0],[1,0,0],[0,0,1]])\n",
        "        perm_matrices[3]=np.array([[0,1,0],[0,0,1],[1,0,0]])\n",
        "        perm_matrices[4]=np.array([[0,0,1],[1,0,0],[0,1,0]])\n",
        "        perm_matrices[5]=np.array([[0,0,1],[0,1,0],[1,0,0]])\n",
        "        perm_matrices=perm_matrices.reshape(6,-1)\n",
        "        if MODE=='TRAIN':\n",
        "            perm=np.arange(TRAIN_SIZE)\n",
        "            while True:\n",
        "                np.random.shuffle(perm)\n",
        "                for i in range(TRAIN_SIZE/BATCH_SIZE):\n",
        "                    train_images,train_sketches=\\\n",
        "                    lib.ut_zap50k.load('TRAIN', DATA_DIR, SKETCH_DIR,[perm[j+i*BATCH_SIZE] for j in range(BATCH_SIZE)])                  \n",
        "                    if PERMUTE_COLORS:\n",
        "                        for j in range(BATCH_SIZE):\n",
        "                            coeffs=np.zeros(COLOR_DIM)\n",
        "                            coeffs[0]=np.random.rand()\n",
        "                            for i in range(COLOR_DIM-1):\n",
        "                                coeffs[i+1]=np.random.rand()*(1-np.sum(coeffs[:(i+1)]))\n",
        "                            coeffs/=np.sum(coeffs)\n",
        "                            np.random.shuffle(perm_matrices)\n",
        "                            ds=np.matmul(coeffs,perm_matrices[:COLOR_DIM]).reshape(3,3)\n",
        "                            train_images[j]=np.matmul(ds,train_images[j].reshape(3,-1)).reshape(3,train_images[j].shape[1],train_images[j].shape[2])    \n",
        "                    yield train_images.reshape((-1,DATA_DIM))*2./255-1\n",
        "        elif MODE=='TEST':\n",
        "            perm=np.arange(TEST_SIZE)\n",
        "            while True:\n",
        "                np.random.shuffle(perm)\n",
        "                for i in range(TEST_SIZE/BATCH_SIZE):\n",
        "                    test_images,test_sketches=\\\n",
        "                    lib.ut_zap50k.load('TEST', DATA_DIR, SKETCH_DIR,[perm[j+i*BATCH_SIZE] for j in range(BATCH_SIZE)])\n",
        "                    yield test_images.reshape((-1,DATA_DIM))*2./255-1\n",
        "                    "
      ],
      "execution_count": 10,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Ik1b9fotdQWh"
      },
      "source": [
        ""
      ],
      "execution_count": 10,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "EZWjirtyJQQg"
      },
      "source": [
        "if not RAND_LABELS:\n",
        "    assert BATCH_SIZE % NUM_CLASSES==0\n",
        "assert NUM_GENERATORS % len(gpu_id)==0"
      ],
      "execution_count": 11,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "87hbTsltJQQj",
        "outputId": "8389399d-9356-44a3-ab00-4b9876cadefb",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 106
        }
      },
      "source": [
        "config=tf.ConfigProto(log_device_placement=True,allow_soft_placement=True)\n",
        "config.graph_options.optimizer_options.global_jit_level = tf.OptimizerOptions.ON_1\n",
        "from tensorflow.core.protobuf import rewriter_config_pb2\n",
        "config.graph_options.rewrite_options.auto_mixed_precision = rewriter_config_pb2.RewriterConfig.ON\n",
        "session=tf.Session(config=config)"
      ],
      "execution_count": 12,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Device mapping:\n",
            "/job:localhost/replica:0/task:0/device:XLA_CPU:0 -> device: XLA_CPU device\n",
            "/job:localhost/replica:0/task:0/device:XLA_GPU:0 -> device: XLA_GPU device\n",
            "/job:localhost/replica:0/task:0/device:GPU:0 -> device: 0, name: Tesla V100-SXM2-16GB, pci bus id: 0000:00:04.0, compute capability: 7.0\n",
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "vTCQY6i1JQQm"
      },
      "source": [
        "#Placeholders\n",
        "with tf.device('/cpu:0'):\n",
        "    real_data = tf.placeholder(dtype, shape=[None, DATA_DIM],name='real_data')\n",
        "    rec_real_data=adv_x=real_data\n",
        "    real_labels = tf.placeholder(tf.int32, shape=[None],name='real_labels')\n",
        "    real_code = tf.one_hot(real_labels,NUM_CLASSES,dtype=dtype)\n",
        "\n",
        "    lr = tf.placeholder(\n",
        "           dtype, [],\n",
        "            name='lr'\n",
        "        )\n",
        "    curr_iter=tf.placeholder(\n",
        "           tf.int32, [],\n",
        "            name='curr_iter'\n",
        "        )\n",
        "    update_p=tf.placeholder(\n",
        "           tf.bool, [],\n",
        "            name='update_p'\n",
        "        )"
      ],
      "execution_count": 13,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "NMKg2ghyJQQr"
      },
      "source": [
        ""
      ],
      "execution_count": 13,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "cuJQTVfXJQQu"
      },
      "source": [
        "#Utilities\n",
        "def mkdir_p(path):\n",
        "    try:\n",
        "        os.makedirs(path)\n",
        "    except OSError as exc:  # Python >2.5\n",
        "        if exc.errno == errno.EEXIST and os.path.isdir(path):\n",
        "            pass\n",
        "        else:\n",
        "            raise\n",
        "# where to load the snapshot, if you need to specify it, comment its subsequent line: #log_dir=''\n",
        "load_dir='/mnt/Data/logs/test'\n",
        "load_dir=''\n",
        "log_dir=os.environ['LOG']+'/MIX-MHingeGAN-%iG%iD-CIFAR-10-conditional-%s/'\\\n",
        "%(NUM_GENERATORS, NUM_DISCRIMINATORS, DATASET)\n",
        "if load_dir=='':\n",
        "    load_dir=log_dir\n",
        "    \n",
        "\n",
        "mkdir_p(log_dir)\n",
        "snapshot_dir = os.path.join(log_dir,\"snapshots/\")#where to save\n",
        "mkdir_p(snapshot_dir)\n",
        "load_snapshot_dir=os.path.join(load_dir,\"snapshots/\")\n",
        "if len([file for file in os.listdir(load_snapshot_dir) if 'ckpt' in file])>0:\n",
        "    g_snapshot_path=os.path.join(load_snapshot_dir, [file for file in os.listdir(load_snapshot_dir) if '_G.ckpt' in file][-1]).split('.ckpt')[0]+'.ckpt'#where to load\n",
        "    d_snapshot_path=g_snapshot_path.replace('_G', '_D')\n",
        "else:\n",
        "    g_snapshot_path=''\n",
        "\n",
        "log_vars=[]\n",
        "def define_summaries():\n",
        "        [g_sum,d_sum,hist_sum,j_sum_train,w_sum_train,j_sum_test,w_sum_test]=[_,_,_,_,_,_,_]\n",
        "        '''Helper function for init_opt'''\n",
        "        all_sum = {'G': [], 'D': [], 'hist': [], 'J_train': [], 'W_train': [], 'J_test': [], 'W_test': []}\n",
        "        for k, v in log_vars:\n",
        "            if k.startswith('G'):\n",
        "                all_sum['G'].append(tf.summary.scalar(k, v))\n",
        "            elif k.startswith('D'):\n",
        "                all_sum['D'].append(tf.summary.scalar(k, v))\n",
        "            elif k.startswith('hist'):\n",
        "                all_sum['hist'].append(tf.summary.histogram(k, v))\n",
        "        if len(all_sum['G'])>0:\n",
        "            g_sum = tf.summary.merge(all_sum['G'])\n",
        "        if len(all_sum['D'])>0:\n",
        "            d_sum = tf.summary.merge(all_sum['D'])\n",
        "        if len(all_sum['hist'])>0:\n",
        "            hist_sum = tf.summary.merge(all_sum['hist'])\n",
        "\n",
        "        return [g_sum,d_sum,hist_sum]\n",
        "\n",
        "\n",
        "def random_pick(some_list, probabilities):  \n",
        "    x = random.uniform(0,1)  \n",
        "    cumulative_probability = 0.0  \n",
        "    for item, item_probability in zip(some_list, probabilities):  \n",
        "        cumulative_probability += item_probability  \n",
        "        if x < cumulative_probability:break  \n",
        "    return item  \n",
        "\n",
        "\n",
        "\n",
        "def label2onehot(labels,num_classes=NUM_CLASSES):\n",
        "    labels=labels.astype(int)\n",
        "    one_hots=np.zeros((len(labels),num_classes))\n",
        "    for i in range(len(labels)):\n",
        "        one_hots[i][labels[i]]=1\n",
        "    return one_hots\n",
        "\n",
        "# For calculating inception score\n",
        "\n",
        "def get_inception_score(n, splits=10):\n",
        "    BATCH_SIZE=int(fake_sample.shape[0])\n",
        "    all_samples = np.zeros([int(np.ceil(float(n)/BATCH_SIZE)*BATCH_SIZE),DATA_DIM],dtype=np.uint8)\n",
        "    if not GET_INCEPTION_SCORE:\n",
        "        return 0,0\n",
        "    labels=np.arange(n)%NUM_CLASSES\n",
        "    for i in range(int(np.ceil(float(n)/BATCH_SIZE))):# inception score for num_batches of fake data\n",
        "        all_samples[i*BATCH_SIZE:(i+1)*BATCH_SIZE]=((\\\n",
        "        session.run(fake_sample,\\\n",
        "        feed_dict={\\\n",
        "        real_labels:labels[i*BATCH_SIZE:(i+1)*BATCH_SIZE].astype(int),\n",
        "        real_code:label2onehot(labels[i*BATCH_SIZE:(i+1)*BATCH_SIZE]),})\\\n",
        "        +1)/2*255).astype(np.uint8)\n",
        "        #all_samples[i*BATCH_SIZE:(i+1)*BATCH_SIZE]=((next(train_gen)[0]+1)/2*255).astype(np.uint8) # IS on real data\n",
        "    return lib.inception_score.get_inception_score(all_samples[:n].reshape([-1,HEIGHT,WIDTH,3]).transpose([0,3,1,2]), splits)\n",
        "def get_fid(n,gen):\n",
        "    if not GET_FID:\n",
        "        return 0\n",
        "    labels=np.arange(n)%NUM_CLASSES\n",
        "    BATCH_SIZE=fake_sample.shape[0]\n",
        "    all_fake_samples = np.zeros([n//BATCH_SIZE*BATCH_SIZE,DATA_DIM],dtype=np.uint8)\n",
        "    for i in range(n//BATCH_SIZE):#\n",
        "        all_fake_samples[i*BATCH_SIZE:(i+1)*BATCH_SIZE]=((session.run(fake_sample,\\\n",
        "        feed_dict={\\\n",
        "        real_labels:labels[i*BATCH_SIZE:(i+1)*BATCH_SIZE].astype(int),\n",
        "        real_code:label2onehot(labels[i*BATCH_SIZE:(i+1)*BATCH_SIZE]),\\\n",
        "        })+1)/2*255).astype(np.uint8)\n",
        "    BATCH_SIZE=next(gen)[0].shape[0]\n",
        "    all_real_samples = np.zeros([n//BATCH_SIZE*BATCH_SIZE,DATA_DIM],dtype=np.uint8)\n",
        "    for i in range(n//BATCH_SIZE):# inception score for num_batches of fake data\n",
        "        all_real_samples[i*BATCH_SIZE:(i+1)*BATCH_SIZE]=((next(gen)[0]+1)/2*255).astype(np.uint8) # range: [0, 255], stored as int8 to save memory\n",
        "    sample_size=min(all_fake_samples.shape[0],all_real_samples.shape[0])\n",
        "    return lib.fid.get_fid(all_real_samples[:sample_size].reshape([-1,HEIGHT,WIDTH,3]).transpose([0,3,1,2]),all_fake_samples[:sample_size].reshape([-1,HEIGHT,WIDTH,3]).transpose([0,3,1,2]))\n",
        "\n",
        "def generate_image(plot_number):\n",
        "        num_images=100 #save 100 images any no matter how big BATCH_SIZE is\n",
        "        num_batches=int(np.ceil(100./BATCH_SIZE)) # number of iterations needed to get 100 images\n",
        "        total_images=num_batches*BATCH_SIZE #might be 128\n",
        "        _labels=np.arange(total_images)//10\n",
        "        _labels[num_images:]=0\n",
        "\n",
        "        _all_samples=[]\n",
        "        _real_data=[]\n",
        "        _rec_real_data=[]\n",
        "        for i in np.arange(num_batches):\n",
        "            real_data_batch,real_code_batch=next(train_gen)\n",
        "            _real_data_batch=session.run(\n",
        "            real_data,\n",
        "            feed_dict={\n",
        "            real_data:real_data_batch\n",
        "            })\n",
        "            #_rec_real_data.append(_rec_real_data_batch)\n",
        "            _real_data.append(_real_data_batch)\n",
        "        for i in np.arange(np.ceil(100./int(fake_sample.shape[0]))):\n",
        "            _samples_batch=session.run(\n",
        "            fake_sample,\n",
        "            feed_dict={})\n",
        "            _all_samples.append(_samples_batch) \n",
        "        _all_samples=np.concatenate(_all_samples)[:100]/2+.5\n",
        "        _real_data=np.concatenate(_real_data)[:100]/2+.5\n",
        "        #_rec_real_data=np.concatenate(_rec_real_data)[:100]/2+.5\n",
        "        \n",
        "        lib.save_images.save_images(\n",
        "        _all_samples.reshape((num_images, HEIGHT, WIDTH, NUM_CHANNELS)), \n",
        "        log_dir+DATASET+'_SAMPLE_'+str(plot_number)+'_all.jpg')\n",
        "        if plot_number % (100000) == 0:\n",
        "            lib.save_images.show_images(_all_samples.reshape((num_images, HEIGHT, WIDTH, NUM_CHANNELS)))\n",
        "        lib.save_images.save_images(\n",
        "        _real_data.reshape((num_images, HEIGHT, WIDTH, NUM_CHANNELS)), \n",
        "        log_dir+DATASET+'_REAL_'+str(plot_number)+'.jpg')\n",
        "        \n",
        "        samples={}\n",
        "        for i in range(NUM_GENERATORS):\n",
        "            samples[i]=[]\n",
        "            for k in range(num_batches):\n",
        "                samples[i].append(session.run(all_fake_sample[i%NUM_GENERATORS]))    \n",
        "            samples[i]=np.concatenate(samples[i])[:100]/2+.5\n",
        "            lib.save_images.save_images(\n",
        "            samples[i].reshape((num_images, HEIGHT, WIDTH, NUM_CHANNELS)), \n",
        "            log_dir+DATASET+'_SAMPLE_'+str(plot_number)+'-'+str(i)+'.jpg')"
      ],
      "execution_count": 14,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ANsB5FzjJQQw"
      },
      "source": [
        "def pixelwise_norm(x, epsilon=1e-8):\n",
        "    return x * tf.rsqrt(tf.reduce_mean(tf.square(x), axis=-1, keepdims=True) + epsilon)"
      ],
      "execution_count": 15,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "JysTUDB-JQQ0",
        "outputId": "50d2d58d-e0ad-489a-f8a9-59a307584426",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        }
      },
      "source": [
        "# Layers and networks definitions\n",
        "\n",
        "if CONDITIONAL and (not ACGAN) and (not NORMALIZATION_D):\n",
        "    print (\"WARNING! Conditional model without normalization in D might be effectively unconditional!\")\n",
        "\n",
        "def nonlinearity(x):\n",
        "    return tf.nn.relu(x)\n",
        "    #return tf.nn.relu6(x)\n",
        "    return tf.nn.leaky_relu(x)\n",
        "def dropout(x, keep_prob):\n",
        "    if DROPOUT:\n",
        "        return tf.nn.dropout(x, keep_prob=keep_prob)\n",
        "    else:\n",
        "        return x\n",
        "def ReLULayer(name, n_in, n_out, inputs):\n",
        "    output = lib.ops.linear.Linear(\n",
        "        name+'.Linear',\n",
        "        n_in,\n",
        "        n_out,\n",
        "        inputs,\n",
        "        initialization='he'\n",
        "    )\n",
        "    output = nonlinearity(output)\n",
        "    return output\n",
        "\n",
        "def Normalize(name, inputs,labels=None, is_training=None):\n",
        "    \"\"\"This is messy, but basically it chooses between batchnorm, layernorm, \n",
        "    their conditional variants, or nothing, depending on the value of `name` and\n",
        "    the global hyperparam flags.\"\"\"\n",
        "    if not CONDITIONAL:\n",
        "        labels = None\n",
        "    if CONDITIONAL and ACGAN and ('Discriminator' in name):\n",
        "        labels = None\n",
        "\n",
        "    if ('Discriminator' in name) and NORMALIZATION_D:\n",
        "        return lib.ops.batchnorm.Batchnorm(name,[0,1,2],inputs,fused=True)\n",
        "        return lib.ops.layernorm.Layernorm(name,[1,2,3],inputs,labels=labels,n_labels=10)\n",
        "    elif ('Generator' in name) and NORMALIZATION_G:\n",
        "        if labels is not None and NUM_GENERATORS < NUM_CLASSES:\n",
        "            return class_condition_batch_norm(name, inputs, labels=labels, is_training=is_training, center=True, scale=True, use_bias=False)\n",
        "            return condition_batch_norm(name, inputs, tf.one_hot(labels, NUM_CLASSES), is_training=is_training, center=True, scale=True, use_bias=False)\n",
        "            return lib.ops.cond_batchnorm.Batchnorm(name,[0,1,2],inputs,labels=labels,n_labels=10)\n",
        "        else:\n",
        "            return batch_norm(name, inputs, is_training=is_training)\n",
        "            return lib.ops.batchnorm.Batchnorm(name,[0,1,2],inputs,fused=True, is_training=True)\n",
        "    else:\n",
        "        return inputs\n",
        "\n",
        "def separable_conv2d(name, input_dim, output_dim, filter_size, inputs, stride=1, labels=None,he_init=True, biases=False):\n",
        "    name=name+'.separable'\n",
        "    #assert(filter_size==3)\n",
        "    output = lib.ops.conv2d.Conv2D(name+'.depthwise',input_dim, input_dim, 3, inputs, depthwise=True, stride=stride,he_init=he_init,biases=biases)\n",
        "    output = Normalize(name+'.BN',output, labels=labels)\n",
        "    output = nonlinearity(output)\n",
        "    output = lib.ops.conv2d.Conv2D(name+'.pointwise', input_dim, output_dim, 1, output,he_init=he_init,biases=biases)  \n",
        "    return output\n",
        "\n",
        "conv2d=functools.partial(lib.ops.conv2d.Conv2D, spectralnorm=SPECTRAL_NORMALIZATION)\n",
        "linear=functools.partial(lib.ops.linear.Linear, spectralnorm=SPECTRAL_NORMALIZATION)\n",
        "\n",
        "def SubpixelConv2D(*args, **kwargs):\n",
        "    kwargs['output_dim'] = 4*kwargs['output_dim']\n",
        "    output = conv2d(*args, **kwargs)\n",
        "    #output = tf.transpose(output, [0,2,3,1])\n",
        "    output = tf.depth_to_space(output, 2, data_format='NHWC')\n",
        "    #output = tf.transpose(output, [0,3,1,2])\n",
        "    return output\n",
        "\n",
        "def BottleneckResidualBlock(name, input_dim, output_dim, filter_size, inputs, resample=None, he_init=True, labels=None):\n",
        "    \"\"\"\n",
        "    resample: None, 'down', or 'up'\n",
        "    \"\"\"\n",
        "    if resample=='down':\n",
        "        conv_shortcut = functools.partial(conv2d, stride=2)\n",
        "        conv_1        = functools.partial(conv2d, input_dim=input_dim, output_dim=input_dim/2)\n",
        "        conv_1b       = functools.partial(conv2d, input_dim=input_dim/2, output_dim=output_dim/2, stride=2)\n",
        "        conv_2        = functools.partial(conv2d, input_dim=output_dim/2, output_dim=output_dim)\n",
        "    elif resample=='up':\n",
        "        conv_shortcut = SubpixelConv2D\n",
        "        conv_1        = functools.partial(conv2d, input_dim=input_dim, output_dim=input_dim/2)\n",
        "        conv_1b       = functools.partial(lib.ops.deconv2d.Deconv2D, input_dim=input_dim/2, output_dim=output_dim/2)\n",
        "        conv_2        = functools.partial(conv2d, input_dim=output_dim/2, output_dim=output_dim)\n",
        "    elif resample==None:\n",
        "        conv_shortcut = conv2d\n",
        "        conv_1        = functools.partial(conv2d, input_dim=input_dim,  output_dim=input_dim/2)\n",
        "        conv_1b       = functools.partial(conv2d, input_dim=input_dim/2,  output_dim=output_dim/2)\n",
        "        conv_2        = functools.partial(conv2d, input_dim=input_dim/2, output_dim=output_dim)\n",
        "    else:\n",
        "        raise Exception('invalid resample value')\n",
        "\n",
        "    if output_dim==input_dim and resample==None:\n",
        "        shortcut = inputs # Identity skip-connection\n",
        "    else:\n",
        "        shortcut = conv_shortcut(name+'.Shortcut', input_dim=input_dim, output_dim=output_dim, filter_size=1,\n",
        "                                 he_init=False, biases=True, inputs=inputs,labels=labels)\n",
        "    output = inputs\n",
        "    output = nonlinearity(output)\n",
        "    output = conv_1(name+'.Conv1', filter_size=1, inputs=output, he_init=he_init,labels=labels)\n",
        "    output = nonlinearity(output)\n",
        "    output = conv_1b(name+'.Conv1B', filter_size=filter_size, inputs=output, he_init=he_init,labels=labels)\n",
        "    output = nonlinearity(output)\n",
        "    output = conv_2(name+'.Conv2', filter_size=1, inputs=output, he_init=he_init, biases=False,labels=labels)\n",
        "    output = Normalize(name+'.BN', [0,2,3], output)\n",
        "    return shortcut + (0.3*output)\n",
        "\n",
        "def ConvMeanPool(name, input_dim, output_dim, filter_size, inputs, he_init=True, biases=True):\n",
        "    output = conv2d(name, input_dim, output_dim, filter_size, inputs, he_init=he_init, biases=biases)\n",
        "    output = tf.add_n([output[:,::2,::2,:], output[:,1::2,::2,:], output[:,::2,1::2,:], output[:,1::2,1::2,:]]) / 4.\n",
        "    return output\n",
        "\n",
        "def MeanPoolConv(name, input_dim, output_dim, filter_size, inputs, he_init=True, biases=True):\n",
        "    output = inputs\n",
        "    output = tf.add_n([output[:,::2,::2,:], output[:,1::2,::2,:], output[:,::2,1::2,:], output[:,1::2,1::2,:]]) / 4.\n",
        "    output = conv2d(name, input_dim, output_dim, filter_size, output, he_init=he_init, biases=biases)\n",
        "    return output\n",
        "\n",
        "def UpsampleConv(name, input_dim, output_dim, filter_size, inputs, he_init=True, biases=True):\n",
        "    output = inputs\n",
        "    #output = tf.concat([output, output, output, output], axis=1)\n",
        "    #output = tf.transpose(output, [0,2,3,1])\n",
        "    #print(output)\n",
        "    output = tf.image.resize_nearest_neighbor(output, [2*int(output.shape[1]), 2*int(output.shape[2])])\n",
        "    #output = tf.image.resize_bilinear(output, [2*int(output.shape[1]), 2*int(output.shape[2])])\n",
        "    #print(output)\n",
        "    #output = tf.depth_to_space(output, 2)\n",
        "    #output = tf.transpose(output, [0,3,1,2])\n",
        "    output = conv2d(name, input_dim, output_dim, filter_size, output, he_init=he_init, biases=biases)\n",
        "    return output\n",
        "\n",
        "def ResidualBlock(name, input_dim, output_dim, filter_size, inputs, resample=None, no_dropout=False, labels=None, is_training=None, wide=True):\n",
        "    \"\"\"\n",
        "    resample: None, 'down', or 'up'\n",
        "    \"\"\"\n",
        "    if resample=='down':\n",
        "        conv_1        = functools.partial(conv2d, input_dim=input_dim, output_dim=output_dim if wide else input_dim)\n",
        "        conv_2        = functools.partial(ConvMeanPool, input_dim=output_dim if wide else input_dim, output_dim=output_dim)\n",
        "        conv_shortcut = ConvMeanPool\n",
        "    elif resample=='up':\n",
        "        conv_1        = functools.partial(UpsampleConv, input_dim=input_dim, output_dim=output_dim)\n",
        "        conv_shortcut = UpsampleConv\n",
        "        conv_2        = functools.partial(conv2d, input_dim=output_dim, output_dim=output_dim)\n",
        "    elif resample==None:\n",
        "        conv_shortcut =conv2d\n",
        "        conv_1        = functools.partial(conv2d, input_dim=input_dim, output_dim=output_dim)\n",
        "        conv_2        = functools.partial(conv2d, input_dim=output_dim, output_dim=output_dim)\n",
        "    else:\n",
        "        raise Exception('invalid resample value')\n",
        "\n",
        "    if output_dim==input_dim and resample==None:\n",
        "        shortcut = inputs # Identity skip-connection\n",
        "    else:\n",
        "        shortcut = conv_shortcut(name+'.Shortcut', input_dim=input_dim, output_dim=output_dim, filter_size=1, he_init=False, biases=True, inputs=inputs)\n",
        "\n",
        "    output = inputs\n",
        "    output = Normalize(name+'.N1', output, labels=labels, is_training=is_training)\n",
        "    output = nonlinearity(output)\n",
        "    if 'tGenerator' in name:\n",
        "        output=pixelwise_norm(output)\n",
        "    output = conv_1(name+'.Conv1', filter_size=filter_size, inputs=output)    \n",
        "    output = Normalize(name+'.N2', output, labels=labels, is_training=is_training)\n",
        "    output = nonlinearity(output)  \n",
        "    if 'tGenerator' in name:\n",
        "        output=pixelwise_norm(output)\n",
        "    output = conv_2(name+'.Conv2', filter_size=filter_size, inputs=output)\n",
        "\n",
        "    return shortcut + output\n",
        "\n",
        "def OptimizedResBlockDisc1(name, inputs):\n",
        "    conv_1        = functools.partial(conv2d, input_dim=3, output_dim=D_DIM)\n",
        "    conv_2        = functools.partial(ConvMeanPool, input_dim=D_DIM, output_dim=D_DIM)\n",
        "    conv_shortcut = MeanPoolConv\n",
        "    shortcut = conv_shortcut(name+'.1.Shortcut', input_dim=3, output_dim=D_DIM, filter_size=1, he_init=False, biases=True, inputs=inputs)\n",
        "\n",
        "    output = inputs\n",
        "    output = conv_1(name+'.1.Conv1', filter_size=3, inputs=output)    \n",
        "    output = nonlinearity(output)            \n",
        "    output = conv_2(name+'.1.Conv2', filter_size=3, inputs=output)\n",
        "    return shortcut + output\n",
        "\n",
        "def _Generator(ID, noise, n_samples, labels=None, ema=False, is_training=True):\n",
        "    name= 'Generator'+'.%i'%ID if not ema else 'Generator_ema'+'.%i'%ID\n",
        "    if noise is None:\n",
        "        noise = tf.random_normal([BATCH_SIZE, Z_DIM],dtype=dtype)\n",
        "    if labels is not None:\n",
        "        assert noise.shape[0]==labels.shape[0], (noise.shape[0],labels.shape[0])\n",
        "    output = linear(name+'.Input', Z_DIM, 4**2*G_DIM, noise, biases=False)\n",
        "    output = tf.reshape(output, [-1, 4, 4, G_DIM])\n",
        "    output = ResidualBlock(name+'.1', G_DIM, G_DIM, 3, output, resample='up', labels=labels, is_training=is_training)\n",
        "    output = ResidualBlock(name+'.2', G_DIM, G_DIM, 3, output, resample='up', labels=labels, is_training=is_training)\n",
        "    output = ResidualBlock(name+'.3', G_DIM, G_DIM, 3, output, resample='up', labels=labels, is_training=is_training)\n",
        "    #output = ResidualBlock(name+'.3', G_DIM, G_DIM, 3, output, resample='up', labels=labels, is_training=is_training)\n",
        "    output = batch_norm(name+'.last_bn',output,is_training=is_training)\n",
        "    output = nonlinearity(output)\n",
        "    output = conv2d(name+'.Output', G_DIM, 3, 3, output, stride=1)\n",
        "    output = tf.tanh(output)\n",
        "    return tf.reshape(output, [-1, DATA_DIM])\n",
        "\n",
        "def MobilenetDiscriminator(\\\n",
        "    ID, inputs, labels,kp1,kp2,kp3, D_DIM=D_DIM, OUT_DIM=1,\\\n",
        "    code=None, NUM_CHANNELS=3, HEIGHT=HEIGHT, WIDTH=WIDTH): #allow code as condition\n",
        "    name='Discriminator%i'%ID\n",
        "    output=tf.reshape(inputs, [-1, 3, HEIGHT, WIDTH]) \n",
        "    output = OptimizedResBlockDisc1(output,ID)\n",
        "    print(output)\n",
        "    D_LAYERS=int(np.ceil(np.log2(HEIGHT/4)))+2  \n",
        "    for i in range(2):# e.g., [32, 16, 8] for CIFAR-10\n",
        "        scale=HEIGHT//2**(i+1)\n",
        "        output=conv2d(name+'.%i'%scale,2**i*D_DIM, 2*2**i*D_DIM, 3, output, stride=2)\n",
        "        output = Normalize(name+'.%i.BN'%scale,output)\n",
        "        output = nonlinearity(output)\n",
        "        for k in range(5 if i==1 else 1):\n",
        "            output = conv2d(name+'.%i.%i'% (scale,k), 2*2**i*D_DIM, 2*2**i*D_DIM, 3, output)\n",
        "            output = Normalize(name+'.%i.%i.BN'%(scale,k),output)\n",
        "            output = nonlinearity(output)\n",
        "    print('before pooling:', output)\n",
        "    output2 = tf.reduce_mean(output, axis=[2,3])  #corresponding to D_\n",
        "    output_wgan = lib.ops.linear.Linear('Discriminator%i.Output'%ID, 2*2**i*D_DIM, 1, output2)\n",
        "    output_wgan = tf.reshape(output_wgan, [-1])    #conrresponding to D\n",
        "    if CONDITIONAL and ACGAN:\n",
        "        output_acgan = lib.ops.linear.Linear('Discriminator%i.ACGANOutput'%ID, 2*2**i*D_DIM, NUM_CLASSES, output2)\n",
        "        return output_wgan, output2, output_acgan\n",
        "    else:\n",
        "        return output_wgan, output2, None  # two layers' of output\n",
        "\n",
        "def MobilenetV1( # untested\n",
        "    ID, inputs, labels,kp1,kp2,kp3, D_DIM=D_DIM, OUT_DIM=1,\\\n",
        "    code=None, NUM_CHANNELS=3, HEIGHT=HEIGHT, WIDTH=WIDTH): #allow code as condition\n",
        "    name='Discriminator%i'%ID\n",
        "    output=tf.reshape(inputs, [BATCH_SIZE, 3, HEIGHT, WIDTH]) \n",
        "    output = OptimizedResBlockDisc1(output,ID)\n",
        "    output = Normalize(name+'.In',output)\n",
        "    output = nonlinearity(output)\n",
        "\n",
        "    D_LAYERS=int(np.ceil(np.log2(HEIGHT/4)))+2  \n",
        "    for i in range(1):# e.g., [32, 16, 8] for CIFAR-10\n",
        "        scale=HEIGHT//2**(i)\n",
        "        output=conv2d(name+'.%i'%scale,D_DIM, D_DIM, 3, output, stride=2)\n",
        "        output = Normalize(name+'.%i.BN'%scale,output)\n",
        "        output = nonlinearity(output)\n",
        "        for k in range(5 if i==0 else 1):\n",
        "            output = conv2d(name+'.%i.%i'% (scale,k), D_DIM, D_DIM, 3, output)\n",
        "            output = Normalize(name+'.%i.%i.BN'%(scale,k),output)\n",
        "            output = nonlinearity(output)\n",
        "    print('before pooling:', output)\n",
        "    output2 = tf.reduce_mean(output, axis=[2,3])  #corresponding to D_\n",
        "    output_wgan = lib.ops.linear.Linear('Discriminator%i.Output'%ID, D_DIM, 1, output2)\n",
        "    output_wgan = tf.reshape(output_wgan, [-1])    #conrresponding to D\n",
        "    if CONDITIONAL and ACGAN:\n",
        "        output_acgan = lib.ops.linear.Linear('Discriminator%i.ACGANOutput'%ID, D_DIM, NUM_CLASSES, output2)\n",
        "        return output_wgan, output2, output_acgan\n",
        "    else:\n",
        "        return output_wgan, output2, None  # two layers' of output\n",
        "    \n",
        "def V2Block(name, input_dim, output_dim, filter_size, inputs, resample=None, relu=False, labels=None, t=6):\n",
        "    \"\"\"\n",
        "    resample: None, 'down', or 'up'\n",
        "    \"\"\"\n",
        "    \"\"\"\n",
        "    resample: None, 'down', or 'up'\n",
        "    \"\"\"\n",
        "    if resample=='down':\n",
        "        conv_1        = functools.partial(conv2d, input_dim=input_dim, output_dim=t*input_dim, stride=1)\n",
        "        conv_2_3        = functools.partial(separable_conv2d, input_dim=t*input_dim, output_dim=output_dim, stride=2)\n",
        "    elif resample==None:\n",
        "        conv_1        = functools.partial(conv2d, input_dim=input_dim, output_dim=t*input_dim, stride=1)\n",
        "        conv_2_3     = functools.partial(separable_conv2d, input_dim=t*input_dim, output_dim=output_dim, stride=1)\n",
        "    else:\n",
        "        raise Exception('invalid resample value')\n",
        "\n",
        "    if resample==None and output_dim==input_dim:\n",
        "        shortcut = inputs # Identity skip-connection\n",
        "    else:\n",
        "        shortcut = None\n",
        "\n",
        "    output = inputs\n",
        "    output = Normalize(name+'.N1', output, labels=labels)\n",
        "    if relu:\n",
        "        output = nonlinearity(output)\n",
        "    output = conv_1(name+'.expand', filter_size=1, inputs=output,labels=labels,he_init=True, biases=False)    \n",
        "    output = Normalize(name+'.N2', output, labels=labels)\n",
        "    output = nonlinearity(output)            \n",
        "    output = conv_2_3(name+'.dwise_n_linear', filter_size=filter_size, inputs=output,labels=labels)\n",
        "    if shortcut is None:\n",
        "        return output\n",
        "    else:\n",
        "        return output + shortcut\n",
        "\n",
        "def V2Blocks_Xn(name, input_dim, output_dim, filter_size, inputs, resample=None, relu=False, labels=None, t=6 ,n=1):\n",
        "    output=inputs\n",
        "    for k in range(n-1):\n",
        "        output= V2Block(name+'.Blk.%i'%k, input_dim, input_dim, filter_size, output, None, False, labels, t)\n",
        "    output=V2Block(name+'.Blk.last', input_dim, output_dim, filter_size, output, resample, relu, labels, t) #rulu only after first conv\n",
        "    return output\n",
        "\n",
        "def MobilenetV2(\\\n",
        "    ID, inputs, labels,kp1,kp2,kp3, D_DIM=D_DIM, OUT_DIM=1,\\\n",
        "    code=None, NUM_CHANNELS=3, HEIGHT=HEIGHT, WIDTH=WIDTH): #allow code as condition\n",
        "    name='Discriminator%i'%ID\n",
        "    output=tf.reshape(inputs, [BATCH_SIZE, 3, HEIGHT, WIDTH]) \n",
        "    print('input:',output)\n",
        "    output = lib.ops.conv2d.Conv2D(name+'.In', 3, D_DIM, 3, output, he_init=False, stride=1)\n",
        "\n",
        "    output = V2Blocks_Xn(name+'.112.1', D_DIM, D_DIM//2, 3, output, resample=None, relu=True, labels=None, t=1, n=1)\n",
        "    output = V2Blocks_Xn(name+'.112.2', D_DIM//2, D_DIM//4*3, 3, output, resample='down', labels=None, t=6, n=2)\n",
        "    output = V2Blocks_Xn(name+'.56', D_DIM//4*3, D_DIM, 3, output, resample='down', labels=None, t=6, n=3)\n",
        "    output = V2Blocks_Xn(name+'.28', D_DIM, 2*D_DIM, 3, output, resample='down', labels=None, t=6, n=4)\n",
        "    output = V2Blocks_Xn(name+'.14.1', 2*D_DIM, 3*D_DIM, 3, output, resample=None, labels=None, t=6, n=3)\n",
        "    output = V2Blocks_Xn(name+'.14.2', 3*D_DIM, 5*D_DIM, 3, output, resample='down', labels=None, t=6, n=3)\n",
        "    output = V2Blocks_Xn(name+'.7.1', 5*D_DIM, 10*D_DIM, 3, output, resample=None, labels=None, t=6, n=1)\n",
        "    output = Normalize(name+'.before_last_conv',output)\n",
        "    output = lib.ops.conv2d.Conv2D(name+'.7.2', 10*D_DIM, 40*D_DIM, 1, output, he_init=True, biases=False, stride=1)\n",
        "    output = Normalize(name+'.before_last_pool',output)\n",
        "    output = nonlinearity(output)\n",
        "    print('before pooling:', output)\n",
        "    output2 = tf.reduce_mean(output, axis=[2,3])  #corresponding to D_\n",
        "    output_wgan = lib.ops.linear.Linear('Discriminator%i.Output'%ID, 1280, 1, output2)\n",
        "    output_wgan = tf.reshape(output_wgan, [-1])    #conrresponding to D\n",
        "    if CONDITIONAL and ACGAN:\n",
        "        output_acgan = lib.ops.linear.Linear('Discriminator%i.ACGANOutput'%ID, 1280, NUM_CLASSES, output2)\n",
        "        return output_wgan, output2, output_acgan\n",
        "    else:\n",
        "        return output_wgan, output2, None  # two layers' of output\n",
        "    \n",
        "\n",
        "\n",
        "def concat_shuffle_split(x, y):\n",
        "        shape = tf.shape(x)\n",
        "        batch_size = shape[0]\n",
        "        height, width = shape[1], shape[2]\n",
        "        depth = x.shape[3].value\n",
        "        z = tf.stack([x, y], axis=3)  # shape [batch_size, height, width, depth, 2]\n",
        "        z = tf.transpose(z, [0, 1, 2, 4, 3])\n",
        "        z = tf.reshape(z, [batch_size, height, width, 2*depth])\n",
        "        x, y = tf.split(z, num_or_size_splits=2, axis=3)\n",
        "        return x, y\n",
        "\n",
        "def basic_unit(name, x):\n",
        "    in_channels = x.shape[3].value\n",
        "    x = lib.ops.conv2d.Conv2D(name+'.pointwise', in_channels,in_channels, 1, x, stride=1)\n",
        "    x = Normalize(name+'.bn1',x)\n",
        "    x = nonlinearity(x)\n",
        "    x = conv2d(name+'.separable', in_channels, in_channels, 3, x, stride=1, nonlinearity=None)\n",
        "    x = Normalize(name+'.bn2',x)\n",
        "    x = nonlinearity(x)\n",
        "    return x\n",
        "\n",
        "def basic_unit_with_downsampling(name, x, out_channels=None):\n",
        "    in_channels = x.shape[3].value\n",
        "    out_channels = 2 * in_channels if out_channels is None else out_channels\n",
        "    \n",
        "    right = lib.ops.conv2d.Conv2D(name+'.pointwise_right', in_channels,in_channels, 1,x, stride=1)\n",
        "    right = Normalize(name+'.bn_right1',right)\n",
        "    right = nonlinearity(right)\n",
        "    right = conv2d(name+'.separable_right', in_channels, out_channels // 2,  1, right, stride=2, nonlinearity=None)  \n",
        "    right = Normalize(name+'.bn_right2',right)\n",
        "    right = nonlinearity(right)\n",
        "    \n",
        "    left = conv2d(name+'.separable_left', in_channels, out_channels // 2,  1, x, stride=2, nonlinearity=None)\n",
        "    left = Normalize(name+'.bn_left',left)\n",
        "    left = nonlinearity(left)\n",
        "    \n",
        "    return left, right\n",
        "\n",
        "def block(name, x, num_units, out_channels=None):\n",
        "    x, y = basic_unit_with_downsampling(name+'.0',x, out_channels)\n",
        "    for j in range(num_units - 1):\n",
        "        x, y = concat_shuffle_split(x, y)\n",
        "        x = basic_unit(name+'.%i'%(j+1), x)\n",
        "    x = tf.concat([x, y], axis=1)\n",
        "    return x\n",
        "\n",
        "def ShufflenetV2( # untested\n",
        "    ID, inputs, labels,kp1,kp2,kp3, D_DIM=D_DIM, OUT_DIM=1,\\\n",
        "    code=None, NUM_CHANNELS=3, HEIGHT=HEIGHT, WIDTH=WIDTH): #allow code as condition\n",
        "    name='Discriminator%i'%ID\n",
        "    output=tf.reshape(inputs, [BATCH_SIZE, HEIGHT, WIDTH, 3]) \n",
        "    print('input:',output)\n",
        "    output=lib.ops.conv2d.Conv2D(name+'.conv1',3, 24, 3, output, stride=1)\n",
        "    output = Normalize(name+'conv1.BN',output)\n",
        "    output = nonlinearity(output)\n",
        "    #output = tf.nn.max_pool(output,ksize=[1,1,3,3],strides=[1,1,2,2],padding='SAME',data_format='NCHW')\n",
        "    complexity = 244  #0.5x: 48, 1x: 116, 1.5x: 176, 2x: 244 \n",
        "    output = block(name+'.Stage2', output, num_units=4, out_channels=complexity) \n",
        "    print('after stage 2', output)\n",
        "    output = block(name+'.Stage3',output, num_units=8)\n",
        "    print('after stage 3', output)\n",
        "    output = lib.ops.conv2d.Conv2D(name+'.conv5', int(output.shape[3]), 1024 if complexity < 244 else 2048, 1, output, he_init=True, biases=False, stride=1)\n",
        "    output = Normalize(name+'.before_last_pool',output)\n",
        "    output = nonlinearity(output)\n",
        "    print('before pooling:', output)\n",
        "    output2 = tf.reduce_mean(output, axis=[1,2])  #corresponding to D_\n",
        "    output_wgan = lib.ops.linear.Linear('Discriminator%i.Output'%ID, 1024 if complexity < 244 else 2048, 1, output2)\n",
        "    output_wgan = tf.reshape(output_wgan, [-1])    #conrresponding to D\n",
        "    if CONDITIONAL and ACGAN:\n",
        "        output_acgan = lib.ops.linear.Linear('Discriminator%i.ACGANOutput'%ID, 1024 if complexity < 244 else 2048, NUM_CLASSES, output2)\n",
        "        return output_wgan, output2, output_acgan\n",
        "    else:\n",
        "        return output_wgan, output2, None  # two layers' of output\n",
        "    \n",
        "    \n",
        "def _Discriminator(ID, inputs, labels,kp1,kp2,kp3, DEVICE_ID=None): # three more parameters of keep rate\n",
        "    name='Discriminator%i'%(ID)\n",
        "    output = tf.reshape(inputs, [-1, HEIGHT, WIDTH, 3])\n",
        "    output = OptimizedResBlockDisc1(name, output)\n",
        "    print(output)\n",
        "    output = ResidualBlock(name+'.3', D_DIM, D_DIM, 3, output, resample='down', labels=labels)\n",
        "    output = dropout(output, keep_prob=kp1)     #dropout after activator\n",
        "    output = ResidualBlock(name+'.4', D_DIM, D_DIM, 3, output, resample=None, labels=labels)\n",
        "    output = dropout(output, keep_prob=kp2)     #dropout after activator\n",
        "    output = ResidualBlock(name+'.5', D_DIM, D_DIM, 3, output, resample=None, labels=labels)\n",
        "    output = dropout(output, keep_prob=kp2)     #dropout after activator\n",
        "    #output = ResidualBlock(name+'.3.0', D_DIM, D_DIM, 3, output, resample='down', labels=labels)\n",
        "    #output = dropout(output, keep_prob=kp1)     #dropout after activator\n",
        "   # output = ResidualBlock(name+'.3.1', D_DIM, D_DIM, 3, output, resample=None, labels=labels)\n",
        "   # output = dropout(output, keep_prob=kp3)     #dropout after activator\n",
        "   # output = ResidualBlock(name+'.3.2', D_DIM, D_DIM, 3, output, resample=None, labels=labels)\n",
        "  #  output = ResidualBlock(name+'.4'.0, D_DIM, D_DIM, 3, output, resample='down', labels=labels)\n",
        "   # output = dropout(output, keep_prob=kp1)     #dropout after activator\n",
        "   # output = ResidualBlock(name+'.4.1', D_DIM, D_DIM, 3, output, resample=None, labels=labels)\n",
        "   # output = dropout(output, keep_prob=kp3)     #dropout after activator\n",
        "   # output = ResidualBlock(name+'.4.2', D_DIM, D_DIM, 3, output, resample=None, labels=labels)\n",
        "    #output = dropout(output, keep_prob=kp3)     #dropout after activator\n",
        "    #output=minibatch_stddev_layer(output)\n",
        "    #output =  lib.ops.conv2d.Conv2D(name+'.refine', \\\n",
        "      #                                     int(output.shape[-1]), int(output.shape[-1])-1,3,output,stride=1)\n",
        "    \n",
        "    output = nonlinearity(output)\n",
        "    output2 =  tf.reduce_sum(output, axis=[1,2])\n",
        "\n",
        "         #  shape = output2.get_shape().as_list() # shape is [BS,1,1,n]\n",
        "            #all_embeddings = lib.get_param(name+'.embed', [NUM_CLASSES, shape[1]], dtype, tf.glorot_uniform_initializer())\n",
        "            #embedding = tf.nn.embedding_lookup(all_embeddings, labels)\n",
        "    embedding = linear(name+'.embed', NUM_CLASSES, int(output2.shape[-1]), tf.one_hot(labels, NUM_CLASSES), biases=False, weight_init=tf.initializers.glorot_normal())\n",
        "    output_wgan = linear('Discriminator%i.Output'%ID, int(output2.shape[-1]), 1, output2)\n",
        "    output_wgan = tf.reshape(output_wgan, [-1])+tf.reduce_sum(embedding*output2, 1)    #conrresponding to D\n",
        "    if CONDITIONAL and ACGAN:\n",
        "        output_acgan =linear('Discriminator%i.ACGANOutput'%ID, int(output2.shape[-1]), NUM_CLASSES, output2)\n",
        "        return output_wgan, output2, output_acgan\n",
        "    else:\n",
        "        return output_wgan, output2, None  # two layers' of output\n",
        "    \n",
        "#Discriminator=ShufflenetV2\n"
      ],
      "execution_count": 16,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "WARNING! Conditional model without normalization in D might be effectively unconditional!\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "UahBnGLtJQQ3"
      },
      "source": [
        "from tensorflow.python.training import moving_averages \n",
        "from absl import logging\n",
        "def _moving_moments_for_inference(name, mean, variance, is_training, decay):\n",
        "  \"\"\"Use moving averages of moments during inference.\n",
        "  Args:\n",
        "    mean: Tensor of shape [num_channels] with the mean of the current batch.\n",
        "    variance: Tensor of shape [num_channels] with the variance of the current\n",
        "      batch.\n",
        "    is_training: Boolean, wheather to construct ops for training or inference\n",
        "      graph.\n",
        "    decay: Decay rate to use for moving averages.\n",
        "  Returns:\n",
        "    Tuple of (mean, variance) to use. This can the same as the inputs.\n",
        "  \"\"\"\n",
        "  # Create the moving average variables and add them to the appropriate\n",
        "  # collections.\n",
        "  variable_collections = [\n",
        "      tf.GraphKeys.MOVING_AVERAGE_VARIABLES,\n",
        "      tf.GraphKeys.MODEL_VARIABLES, tf.GraphKeys.GLOBAL_VARIABLES,\n",
        "  ]\n",
        "  # Disable partition setting for moving_mean and moving_variance\n",
        "  # as assign_moving_average op below doesn\"t support partitioned variable.\n",
        "  moving_mean = lib.get_param(\n",
        "      name+\".moving_mean\",\n",
        "      shape=mean.shape,\n",
        "      initializer=tf.zeros_initializer(),\n",
        "      trainable=False,\n",
        "      partitioner=None,\n",
        "      )\n",
        "  moving_variance = lib.get_param(\n",
        "      name+\".moving_variance\",\n",
        "      shape=variance.shape,\n",
        "      initializer=tf.ones_initializer(),\n",
        "      trainable=False,\n",
        "      partitioner=None,\n",
        "      )\n",
        "  if is_training:\n",
        "    #logging.debug(\"Adding update ops for moving averages of mean and variance.\")\n",
        "    # Update variables for mean and variance during training.\n",
        "    update_moving_mean = moving_averages.assign_moving_average(\n",
        "        moving_mean,\n",
        "        tf.cast(mean, moving_mean.dtype),\n",
        "        decay,\n",
        "        zero_debias=False)\n",
        "    update_moving_variance = moving_averages.assign_moving_average(\n",
        "        moving_variance,\n",
        "        tf.cast(variance, moving_variance.dtype),\n",
        "        decay,\n",
        "        zero_debias=False)\n",
        "    tf.add_to_collection(tf.GraphKeys.UPDATE_OPS, update_moving_mean)\n",
        "    tf.add_to_collection(tf.GraphKeys.UPDATE_OPS, update_moving_variance)\n",
        "    return mean, variance\n",
        "  #logging.debug(\"Using moving mean and variance.\")\n",
        "  return moving_mean, moving_variance\n",
        "clear_op=[]\n",
        "accu_op=[]\n",
        "def _accumulated_moments_for_inference(name, mean, variance, is_training):\n",
        "  \"\"\"Use accumulated statistics for moments during inference.\n",
        "  After training the user is responsible for filling the accumulators with the\n",
        "  actual values. See _UpdateBnAccumulators() in eval_gan_lib.py for an example.\n",
        "  Args:\n",
        "    mean: Tensor of shape [num_channels] with the mean of the current batch.\n",
        "    variance: Tensor of shape [num_channels] with the variance of the current\n",
        "      batch.\n",
        "    is_training: Boolean, wheather to construct ops for training or inference\n",
        "      graph.\n",
        "  Returns:\n",
        "    Tuple of (mean, variance) to use. This can the same as the inputs.\n",
        "  \"\"\"\n",
        "  #variable_collections = [\n",
        "  #    tf.GraphKeys.MODEL_VARIABLES, tf.GraphKeys.GLOBAL_VARIABLES,\n",
        "  #]\n",
        "  if 1:\n",
        "    # Create variables for accumulating batch statistic and use them during\n",
        "    # inference. The ops for filling the accumulators must be created and run\n",
        "    # before eval. See docstring above.\n",
        "    mean = tf.identity(mean, \"mean\")\n",
        "    variance = tf.identity(variance, \"variance\")\n",
        "    if is_training or not ACCU_STATS:\n",
        "      return mean, variance\n",
        "\n",
        "    accu_mean = lib.get_param(\n",
        "        name+\".accu_mean\",\n",
        "        shape=mean.shape,\n",
        "        initializer=tf.zeros_initializer(),\n",
        "        trainable=False\n",
        "        )\n",
        "    accu_variance = lib.get_param(\n",
        "        name+\".accu_variance\",\n",
        "        shape=variance.shape,\n",
        "        initializer=tf.zeros_initializer(),\n",
        "        trainable=False)\n",
        "    accu_counter = lib.get_param(\n",
        "        name+\".accu_counter\",\n",
        "        shape=[],\n",
        "        initializer=tf.initializers.constant(1e-12),\n",
        "        trainable=False)\n",
        "    update_accus = lib.get_param(\n",
        "        name+\".update_accus\",\n",
        "        shape=[],\n",
        "        dtype=tf.int32,\n",
        "        initializer=tf.ones_initializer(),\n",
        "        trainable=False\n",
        "        )\n",
        "\n",
        "    #logging.debug(\"Using accumulated moments.\")\n",
        "    # Return the accumulated batch statistics and add current batch statistics\n",
        "    # to accumulators if update_accus variables equals 1.\n",
        "    def update_accus_fn():\n",
        "      return tf.group([\n",
        "          tf.assign_add(accu_mean, mean),\n",
        "          tf.assign_add(accu_variance, variance),\n",
        "          tf.assign_add(accu_counter, 1),\n",
        "      ])\n",
        "    dep = tf.cond(\n",
        "        tf.equal(update_accus, 1),\n",
        "        update_accus_fn,\n",
        "        tf.no_op)\n",
        "    clear_op.append(tf.group(*[\n",
        "          tf.assign(accu_mean, 0*accu_mean),\n",
        "          tf.assign(accu_variance, 0*accu_variance),\n",
        "          tf.assign(accu_counter, 1e-12)\n",
        "      ]))\n",
        " \n",
        "    with tf.control_dependencies([dep]):\n",
        "      return accu_mean / accu_counter, accu_variance / accu_counter\n",
        "\n",
        "def standardize_batch(name, inputs,\n",
        "                      is_training,\n",
        "                      decay=0.9,\n",
        "                      epsilon=1e-5,\n",
        "                      data_format=\"NHWC\",\n",
        "                      use_moving_averages=False,\n",
        "                      use_cross_replica_mean=False):\n",
        "  \"\"\"Adds TPU-enabled batch normalization layer.\n",
        "  This version does not apply trainable scale or offset!\n",
        "  It normalizes a tensor by mean and variance.\n",
        "  Details on Batch Normalization can be found in \"Batch Normalization:\n",
        "  Accelerating Deep Network Training by Reducing Internal Covariate Shift\",\n",
        "  Ioffe S. and Szegedy C. 2015 [http://arxiv.org/abs/1502.03167].\n",
        "  Note #1: This method computes the batch statistic across all TPU replicas,\n",
        "  thus simulating the true batch norm in the distributed setting. If one wants\n",
        "  to avoid the cross-replica communication set use_cross_replica_mean=False.\n",
        "  Note #2: When is_training is True the moving_mean and moving_variance need\n",
        "  to be updated in each training step. By default, the update_ops are placed\n",
        "  in `tf.GraphKeys.UPDATE_OPS` and they need to be added as a dependency to\n",
        "  the `train_op`. For example:\n",
        "    update_ops = tf.get_collection(tf.GraphKeys.UPDATE_OPS)\n",
        "    if update_ops:\n",
        "      updates = tf.group(*update_ops)\n",
        "      total_loss = control_flow_ops.with_dependencies([updates], total_loss)\n",
        "  Note #3: Reasonable values for `decay` are close to 1.0, typically in the\n",
        "  multiple-nines range: 0.999, 0.99, 0.9, etc. Lower the `decay` value (trying\n",
        "  `decay`=0.9) if model experiences reasonably good training performance but\n",
        "  poor validation and/or test performance.\n",
        "  Args:\n",
        "    inputs: A tensor with 2 or 4 dimensions, where the first dimension is\n",
        "      `batch_size`. The normalization is over all but the last dimension if\n",
        "      `data_format` is `NHWC`, and the second dimension if `data_format` is\n",
        "      `NCHW`.\n",
        "    is_training: Whether or not the layer is in training mode. In training\n",
        "      mode it would accumulate the statistics of the moments into the\n",
        "      `moving_mean` and `moving_variance` using an exponential moving average\n",
        "      with the given `decay`. When is_training=False, these variables are not\n",
        "      updated, and the precomputed values are used verbatim.\n",
        "    decay: Decay for the moving averages. See notes above for reasonable\n",
        "      values.\n",
        "    epsilon: Small float added to variance to avoid dividing by zero.\n",
        "    data_format: Input data format. NHWC or NCHW.\n",
        "    use_moving_averages: If True keep moving averages of mean and variance that\n",
        "      are used during inference. Otherwise use accumlators.\n",
        "    use_cross_replica_mean: If True add operations to do computes batch norm\n",
        "      statistics across all TPU cores. These ops are not compatible with other\n",
        "      platforms. The default (None) will only add the operations if running\n",
        "      on TPU.\n",
        "  Returns:\n",
        "    The normalized tensor with the same type and shape as `inputs`.\n",
        "  \"\"\"\n",
        "  if data_format not in {\"NCHW\", \"NHWC\"}:\n",
        "    raise ValueError(\n",
        "        \"Invalid data_format {}. Allowed: NCHW, NHWC.\".format(data_format))\n",
        "  if use_cross_replica_mean is None:\n",
        "    # Default to global batch norm only on TPUs.\n",
        "    use_cross_replica_mean = (\n",
        "        tpu_function.get_tpu_context().number_of_shards is not None)\n",
        "    logging.debug(\"Automatically determined use_cross_replica_mean=%s.\",\n",
        "                  use_cross_replica_mean)\n",
        "\n",
        "  inputs = tf.convert_to_tensor(inputs)\n",
        "  inputs_dtype = inputs.dtype\n",
        "  inputs_shape = inputs.get_shape()\n",
        "\n",
        "  num_channels = inputs.shape[-1].value\n",
        "  if num_channels is None:\n",
        "    raise ValueError(\"`C` dimension must be known but is None\")\n",
        "\n",
        "  inputs_rank = inputs_shape.ndims\n",
        "  if inputs_rank is None:\n",
        "    raise ValueError(\"Inputs %s has undefined rank\" % inputs.name)\n",
        "  elif inputs_rank not in [2, 4]:\n",
        "    raise ValueError(\n",
        "        \"Inputs %s has unsupported rank.\"\n",
        "        \" Expected 2 or 4 but got %d\" % (inputs.name, inputs_rank))\n",
        "  # Bring 2-D inputs into 4-D format.\n",
        "  if inputs_rank == 2:\n",
        "    new_shape = [-1, 1, 1, num_channels]\n",
        "    if data_format == \"NCHW\":\n",
        "      new_shape = [-1, num_channels, 1, 1]\n",
        "    inputs = tf.reshape(inputs, new_shape)\n",
        "\n",
        "  # Execute a distributed batch normalization\n",
        "  axis = 1 if data_format == \"NCHW\" else 3\n",
        "  inputs = tf.cast(inputs, tf.float32)\n",
        "  reduction_axes = [i for i in range(4) if i != axis]\n",
        "  if use_cross_replica_mean:\n",
        "    mean, variance = tpu_ops.cross_replica_moments(inputs, reduction_axes)\n",
        "  else:\n",
        "    counts, mean_ss, variance_ss, _ = tf.nn.sufficient_statistics(\n",
        "        inputs, reduction_axes, keep_dims=False)\n",
        "    mean, variance = tf.nn.normalize_moments(\n",
        "        counts, mean_ss, variance_ss, shift=None)\n",
        "  use_moving_averages= 0\n",
        "  if use_moving_averages:\n",
        "    mean, variance = _moving_moments_for_inference(\n",
        "        name, mean=mean, variance=variance, is_training=is_training, decay=decay)\n",
        "  else:\n",
        "    mean, variance = _accumulated_moments_for_inference(\n",
        "        name, mean=mean, variance=variance, is_training=is_training)\n",
        "\n",
        "  outputs = tf.nn.batch_normalization(\n",
        "      inputs,\n",
        "      mean=mean,\n",
        "      variance=variance,\n",
        "      offset=None,\n",
        "      scale=None,\n",
        "      variance_epsilon=epsilon)\n",
        "  outputs = tf.cast(outputs, inputs_dtype)\n",
        "\n",
        "  # Bring 2-D inputs back into 2-D format.\n",
        "  if inputs_rank == 2:\n",
        "    outputs = tf.reshape(outputs, [-1] + inputs_shape[1:].as_list())\n",
        "  outputs.set_shape(inputs_shape)\n",
        "  return outputs\n",
        "'''\n",
        "def condition_batch_norm(name, x, z, is_training=True):\n",
        "        _, _, _ , c= x.get_shape().as_list()\n",
        "        decay = 0.999\n",
        "        epsilon = 1e-4\n",
        "\n",
        "        test_mean = lib.param(name+\".pop_mean\", tf.zeros([1,1,1,c],dtype=dtype), dtype=dtype, trainable=False)\n",
        "        test_var = lib.param(name+\".pop_var\", tf.ones([1,1,1,c],dtype=dtype), dtype=dtype, trainable=False)\n",
        "        beta = linear(name+'.beta', int(z.shape[-1]), c, z, biases=False)\n",
        "        gamma = linear(name+'.gamma', int(z.shape[-1]), c, z, biases=False) + 1\n",
        "\n",
        "        beta = tf.reshape(beta, shape=[-1, 1, 1, c])\n",
        "        gamma = tf.reshape(gamma, shape=[-1, 1, 1, c])\n",
        "\n",
        "        if is_training:\n",
        "            batch_mean, batch_var = tf.nn.moments(x, [0, 1, 2], keep_dims=True)\n",
        "            ema_mean = tf.assign(test_mean, test_mean * decay + batch_mean * (1 - decay))\n",
        "            ema_var = tf.assign(test_var, test_var * decay + batch_var * (1 - decay))\n",
        "\n",
        "            with tf.control_dependencies([ema_mean, ema_var]):\n",
        "                return gamma*tf.nn.batch_normalization(x, test_mean, test_var, offset=None, scale=None,variance_epsilon=epsilon)+beta\n",
        "        else:\n",
        "            return tf.nn.batch_normalization(x, test_mean, test_var, beta, gamma, epsilon)\n",
        "'''\n",
        "def batch_norm(name, inputs, is_training, center=True, scale=True):\n",
        "  \"\"\"Performs the vanilla batch normalization with trainable scaling and offset.\n",
        "  Args:\n",
        "    inputs: A tensor with 2 or 4 dimensions, where the first dimension is\n",
        "      `batch_size`. The normalization is over all but the last dimension if\n",
        "      `data_format` is `NHWC`, and the second dimension if `data_format` is\n",
        "      `NCHW`.\n",
        "    is_training: Whether or not the layer is in training mode.\n",
        "    center: If True, add offset of beta to normalized tensor.\n",
        "    scale: If True, multiply by gamma. When the next layer is linear  this can\n",
        "      be disabled since the scaling will be done by the next layer.\n",
        "    name: Name of the variable scope.\n",
        "  Returns:\n",
        "    The normalized tensor with the same type and shape as `inputs`.\n",
        "  \"\"\"\n",
        "  if 1:\n",
        "    outputs = standardize_batch(name, inputs, is_training=is_training)\n",
        "    num_channels = inputs.shape[-1].value\n",
        "\n",
        "    # Allocate parameters for the trainable variables.\n",
        "#    collections = [tf.GraphKeys.MODEL_VARIABLES,\n",
        "             #      tf.GraphKeys.GLOBAL_VARIABLES]\n",
        "    if scale:\n",
        "      gamma = lib.get_param(\n",
        "          name+\".gamma\",\n",
        "          [num_channels],\n",
        "          initializer=tf.ones_initializer())\n",
        "      outputs *= gamma\n",
        "    if center:\n",
        "      beta = lib.get_param(\n",
        "          name+\".beta\",\n",
        "          [num_channels],\n",
        "          initializer=tf.zeros_initializer())\n",
        "      outputs += beta\n",
        "    return outputs\n",
        "\n",
        "\n",
        "def condition_batch_norm(name, inputs, y, is_training, center=True,\n",
        "                           scale=True, use_bias=False):\n",
        "  \"\"\"Conditional batch normalization.\"\"\"\n",
        "  if y is None:\n",
        "    raise ValueError(\"You must provide y for conditional batch normalization.\")\n",
        "  if y.shape.ndims != 2:\n",
        "    raise ValueError(\"Conditioning must have rank 2.\")\n",
        "  if 1:\n",
        "    outputs = standardize_batch(name, inputs, is_training=is_training)\n",
        "    num_channels = inputs.shape[-1].value\n",
        "    if 1:\n",
        "      if scale:\n",
        "        gamma = linear(name+'.Gamma', int(y.shape[-1]), num_channels, y, biases=use_bias)\n",
        "        gamma = tf.reshape(gamma, [-1, 1, 1, num_channels])\n",
        "        outputs *= gamma\n",
        "      if center:\n",
        "        beta = linear(name+'.Beta', int(y.shape[-1]), num_channels, y, biases=use_bias)\n",
        "        beta = tf.reshape(beta, [-1, 1, 1, num_channels])\n",
        "        outputs += beta\n",
        "      return outputs\n",
        "\n",
        "def class_condition_batch_norm(name, inputs, y, is_training, center=True,\n",
        "                           scale=True, use_bias=False):\n",
        "  \"\"\"Conditional batch normalization.\"\"\"\n",
        "  if y is None:\n",
        "    raise ValueError(\"You must provide labels for conditional batch normalization.\")\n",
        "  if y.shape.ndims != 1:\n",
        "    raise ValueError(\"Labels must have rank 1.\")\n",
        "  if 1:\n",
        "    outputs = standardize_batch(name, inputs, is_training=is_training)\n",
        "    num_channels = inputs.shape[-1].value\n",
        "    if 1:\n",
        "      offset_m = lib.get_param(name+'.offset', [NUM_CLASSES,num_channels], initializer=tf.zeros_initializer())\n",
        "      scale_m = lib.get_param(name+'.scale', [NUM_CLASSES,num_channels], initializer=tf.ones_initializer())\n",
        "      gamma = tf.nn.embedding_lookup(offset_m, y)\n",
        "      beta = tf.nn.embedding_lookup(scale_m, y)\n",
        "      if scale:\n",
        "        gamma = tf.reshape(gamma, [-1, 1, 1, num_channels])\n",
        "        outputs *= gamma\n",
        "      if center:\n",
        "        beta = tf.reshape(beta, [-1, 1, 1, num_channels])\n",
        "        outputs += beta\n",
        "      return outputs\n",
        "\n",
        "def upsample(x):\n",
        "            x = tf.image.resize_nearest_neighbor(x, [2*int(x.shape[1]), 2*int(x.shape[2])])\n",
        "            return x    \n",
        "##################################################################################\n",
        "# Residual-block, Self-Attention-block\n",
        "##################################################################################\n",
        "\n",
        "def GBlock(name, x, y, channels, is_training, use_bias=True, sn=False,use_upsample=True):\n",
        "\n",
        "    in_channels, out_channels = int(x.shape[3]), channels\n",
        "    activation = nonlinearity\n",
        "    upsample_fn = upsample\n",
        "    learnable_sc = in_channels != out_channels or upsample\n",
        "    \n",
        "    # Conv layers\n",
        "    conv1 = functools.partial(conv2d, name=name+'.conv1', input_dim=in_channels, output_dim=out_channels, biases=use_bias, filter_size=3)\n",
        "    conv2 =functools.partial(conv2d, name=name+'.conv2', input_dim=out_channels, output_dim=out_channels, biases=use_bias, filter_size=3)\n",
        "    if learnable_sc:\n",
        "      conv_sc = functools.partial(conv2d, name=name+'.conv_sc',input_dim=in_channels, output_dim=out_channels, biases=use_bias, filter_size=1)\n",
        "    # Batchnorm layers\n",
        "    if 1:\n",
        "        which_bn = condition_batch_norm\n",
        "    else:\n",
        "        which_bn = class_condition_batch_norm\n",
        "    bn = functools.partial(which_bn,is_training=is_training, use_bias=False)\n",
        "\n",
        "    h = activation(bn(name=name+'.cbn1', inputs=x, y=y))\n",
        "    if use_upsample:\n",
        "      h = upsample_fn(h)\n",
        "      x = upsample_fn(x)\n",
        "    h = conv1(inputs=h)\n",
        "    h = activation(bn(name=name+'.cbn2', inputs=h, y=y))\n",
        "    h = conv2(inputs=h)\n",
        "    if learnable_sc:       \n",
        "      x = conv_sc(inputs=x)\n",
        "    return h + x\n",
        "\n",
        "def DBlock(name, x, channels, use_bias=True, is_training=True, sn=False,preactivation=True,downsample=True,wide=True):\n",
        "    '''        #x = Normalize(name+'BN1',x_init)\n",
        "            if preactivation:\n",
        "                x = nonlinearity(x_init)\n",
        "            else:\n",
        "                x = x_init\n",
        "            x = conv2d(name+'res1',int(x.shape[3]), channels, 3, x, stride=1, biases=use_bias)\n",
        "            #x = Normalize(name+'BN2',x)\n",
        "            x = nonlinearity(x)\n",
        "            x = conv2d(name+'res2',int(x.shape[3]), channels, 3, x, stride=1, biases=use_bias)\n",
        "            x=tf.nn.avg_pool(x,ksize=[1,2,2,1],strides=[1,2,2,1],padding='SAME',data_format='NHWC')\n",
        "            if preactivation:\n",
        "                x_init = conv2d(name+'skip',int(x_init.shape[3]), channels, 1, x_init, stride=1,  biases=use_bias)\n",
        "                x_init=tf.nn.avg_pool(x_init,ksize=[1,2,2,1],strides=[1,2,2,1],padding='SAME',data_format='NHWC') \n",
        "            else:\n",
        "                x_init=tf.nn.avg_pool(x_init,ksize=[1,2,2,1],strides=[1,2,2,1],padding='SAME',data_format='NHWC') \n",
        "                x_init = conv2d(name+'skip',int(x_init.shape[3]), channels, 1, x_init, stride=1,  biases=use_bias)\n",
        "            return x + x_init\n",
        "'''\n",
        "    in_channels, out_channels = int(x.shape[3]), channels\n",
        "    # If using wide D (as in SA-GAN and BigGAN), change the channel pattern\n",
        "    hidden_channels = out_channels if wide else in_channels\n",
        "    activation = nonlinearity\n",
        "    downsample_fn=functools.partial(tf.nn.avg_pool,ksize=[1,2,2,1],strides=[1,2,2,1],padding='SAME',data_format='NHWC')\n",
        "    # Conv layers\n",
        "    conv1 = functools.partial(conv2d, name=name+'.conv1', input_dim=in_channels, output_dim=hidden_channels, biases=use_bias, filter_size=3)\n",
        "    conv2 =functools.partial(conv2d, name=name+'.conv2', input_dim=hidden_channels, output_dim=out_channels, biases=use_bias, filter_size=3)\n",
        "    learnable_sc = True if (in_channels != out_channels) or downsample else False\n",
        "    if learnable_sc:\n",
        "        conv_sc=functools.partial(conv2d, name=name+'.conv_sc',input_dim=in_channels, output_dim=out_channels, biases=use_bias, filter_size=1)\n",
        "    def shortcut(x):\n",
        "        if preactivation:\n",
        "          if learnable_sc:\n",
        "            x = conv_sc(inputs=x)\n",
        "          if downsample:\n",
        "            x = downsample_fn(x)\n",
        "        else:\n",
        "          if downsample:\n",
        "            x = downsample_fn(x)\n",
        "          if learnable_sc:\n",
        "            x = conv_sc(inputs=x)\n",
        "        return x\n",
        "    if preactivation:\n",
        "      # h = self.activation(x) # NOT TODAY SATAN\n",
        "      # Andy's note: This line *must* be an out-of-place ReLU or it \n",
        "      #              will negatively affect the shortcut connection.\n",
        "      h = nonlinearity(x)\n",
        "    else:\n",
        "      h = x    \n",
        "    h = conv1(inputs=h)\n",
        "    h = conv2(inputs=nonlinearity(h))\n",
        "    if downsample:\n",
        "      h =downsample_fn(h)\n",
        "    return h + shortcut(x)\n",
        "\n",
        "def bottleneck_resblock(name, x, z, channels, use_bias=True, is_training=True, sn=False):\n",
        "            x_init = x\n",
        "            if 'Generator' in name:\n",
        "                x = condition_batch_norm(name+'.cbn1',x, z, is_training)\n",
        "            x = nonlinearity(x)\n",
        "            x = conv2d(name+'.res1',int(x.shape[3]), int(x.shape[3])//4, 1, x, stride=1, biases=use_bias)\n",
        "            if 'Generator' in name:\n",
        "                x = condition_batch_norm(name+'.cbn2',x, z, is_training)\n",
        "            x = nonlinearity(x_init)\n",
        "            x = conv2d(name+'.res2',int(x.shape[3]), int(x.shape[3]), 3, x, stride=1, biases=use_bias)\n",
        "            if 'Generator' in name:\n",
        "                x = condition_batch_norm(name+'.cbn3',x, z, is_training)\n",
        "            x = nonlinearity(x)\n",
        "            x = conv2d(name+'.res3',int(x.shape[3]), int(x.shape[3]), 3, x, stride=1, biases=use_bias)\n",
        "            if 'Generator' in name:\n",
        "                x = condition_batch_norm(name+'.cbn4',x, z, is_training)\n",
        "            x = nonlinearity(x)\n",
        "            x = conv2d(name+'.res4',int(x.shape[3]), channels, 1, x, stride=1, biases=use_bias)\n",
        "            \n",
        "            return x + x_init\n",
        "        \n",
        "def bottleneck_resblock_up_condition(name, x_init, z, channels, use_bias=True, is_training=True, sn=False):\n",
        "            x = condition_batch_norm(name+'.cbn1',x_init, z, is_training)\n",
        "            x = nonlinearity(x)\n",
        "            x = conv2d(name+'.res1',int(x.shape[3]), int(x.shape[3])//4, 1, x, stride=1, biases=use_bias)\n",
        "            x = condition_batch_norm(name+'.cbn2',x, z, is_training)\n",
        "            x = nonlinearity(x_init)\n",
        "            x = upsample(x)\n",
        "            x = conv2d(name+'.res2',int(x.shape[3]), int(x.shape[3]), 3, x, stride=1, biases=use_bias)\n",
        "            x = condition_batch_norm(name+'.cbn3',x, z, is_training)\n",
        "            x = nonlinearity(x)\n",
        "            x = conv2d(name+'.res3',int(x.shape[3]), int(x.shape[3]), 3, x, stride=1, biases=use_bias)\n",
        "            x = condition_batch_norm(name+'.cbn4',x, z, is_training)\n",
        "            x = nonlinearity(x)\n",
        "            x = conv2d(name+'.res4',int(x.shape[3]), channels, 1, x, stride=1, biases=use_bias)\n",
        "            \n",
        "            x_skip=x_init[:,:,:,:channels]\n",
        "            x_skip=upsample(x_skip)\n",
        "            \n",
        "            return x + x_skip\n",
        "\n",
        "def bottleneck_resblock_down(name, x_init, channels, use_bias=True, is_training=True, sn=False, scope='resblock_down'):\n",
        "            x = nonlinearity(x_init)\n",
        "            x = conv2d(name+'.res1',int(x.shape[3]), int(x.shape[3])//4, 1, x, stride=1, biases=use_bias)\n",
        "            x = nonlinearity(x_init)\n",
        "            x = conv2d(name+'.res2',int(x.shape[3]), int(x.shape[3]), 3, x, stride=1, biases=use_bias)\n",
        "            x = nonlinearity(x)\n",
        "            x = conv2d(name+'.res3',int(x.shape[3]), int(x.shape[3]), 3, x, stride=1, biases=use_bias)\n",
        "            x = nonlinearity(x)\n",
        "            x=tf.nn.avg_pool(x,ksize=[1,2,2,1],strides=[1,2,2,1],padding='SAME',data_format='NHWC')\n",
        "            x = conv2d(name+'.res4',int(x.shape[3]), channels, 1, x, stride=1, biases=use_bias)\n",
        "            x_skip=tf.nn.avg_pool(x_init,ksize=[1,2,2,1],strides=[1,2,2,1],padding='SAME',data_format='NHWC')\n",
        "            if channels>int(x_skip.shape[3]):\n",
        "                x_skip=tf.concat([x_skip, conv2d(name+'.skip',int(x_skip.shape[3]), int(x_skip.shape[3]), 1, x_skip, stride=1,  biases=use_bias)],3)\n",
        "\n",
        "            return x + x_skip\n",
        "        \n",
        "def hw_flatten(x) :\n",
        "    return tf.reshape(x, [-1, int(x.shape[1])*int(x.shape[2]), int(x.shape[3])])\n",
        "\n",
        "def self_attention(name, x, channels, sn=False, scope='self_attention'):\n",
        "\n",
        "        f = conv2d(name+'.sa.conv1',int(x.shape[3]), channels // 8, 1, x, stride=1, sn=sn, scope='f_conv')  # [bs, h, w, c']\n",
        "        g = conv2d(name+'.sa.conv2',int(x.shape[3]), channels // 8, 1, x, stride=1, sn=sn, scope='g_conv')  # [bs, h, w, c']\n",
        "        h = conv2d(name+'.sa.conv3',int(x.shape[3]), channels, 1, x, stride=1, sn=sn, scope='h_conv')  # [bs, h, w, c]\n",
        "\n",
        "        # N = h * w\n",
        "        s = tf.matmul(hw_flatten(g), hw_flatten(f),  transpose_b=True)  # # [bs, N, N]\n",
        "\n",
        "        beta = tf.nn.softmax(s)  # attention map\n",
        "\n",
        "        o = tf.matmul(beta, hw_flatten(h))  # [bs, N, C]\n",
        "        gamma = lib.param(name+\".gamma\", 0.,dtype=dtype)\n",
        "\n",
        "        o = tf.reshape(o, shape=x.shape)  # [bs, h, w, C]\n",
        "        x = gamma * o + x\n",
        "\n",
        "        return x\n",
        "\n",
        "def self_attention_2(name, x, channels, sn=False, biases=False):\n",
        "    \n",
        "        f = conv2d(name+'.sa.conv1',int(x.shape[3]), channels // 8, 1, x, stride=1, biases=biases)  # [bs, h, w, c/8]\n",
        "        f =  tf.nn.max_pool(f,ksize=[1,2,2,1],strides=[1,2,2,1],padding='SAME',data_format='NHWC')# [bs, h/2, w/2, c/8]\n",
        "\n",
        "        g = conv2d(name+'.sa.conv2',int(x.shape[3]), channels // 8, 1, x, stride=1, biases=biases)  # [bs, h, w, c/8]\n",
        "\n",
        "        h = conv2d(name+'.sa.conv3',int(x.shape[3]), channels // 2, 1, x, stride=1, biases=biases)  # [bs, h, w, c/2]\n",
        "        h = tf.nn.max_pool(h,ksize=[1,2,2,1],strides=[1,2,2,1],padding='SAME',data_format='NHWC')# [bs, h/2, w/2, c/2]\n",
        "\n",
        "        # N = h * w\n",
        "        s = tf.matmul(hw_flatten(g), hw_flatten(f),  transpose_b=True)  # # [bs, N, N/4]\n",
        "\n",
        "        beta = tf.nn.softmax(s)  # attention map # [bs, N, N/4]\n",
        "\n",
        "        o = tf.matmul(beta, hw_flatten(h))  # [bs, N, C/2]\n",
        "        gamma = lib.param(name+\".gamma\", 0.,dtype=dtype)\n",
        "\n",
        "        o = tf.reshape(o, shape=[-1, x.shape[1], x.shape[2], channels // 2])  # [bs, h, w, C/2]\n",
        "        o = conv2d(name+'.out',channels // 2, channels, 1, o, stride=1, biases=biases)  # [bs, h, w, C]\n",
        "        x = gamma * o + x\n",
        "\n",
        "        return x\n",
        "\n",
        "    ##################################################################################\n",
        "    # Generator\n",
        "    ##################################################################################\n",
        "\n",
        "def Generator(ID, noise, n_samples, labels=None, ema=False, is_training=True):\n",
        "            name= 'Generator'+'.%i'%ID if not ema else 'Generator_ema'+'.%i'%ID\n",
        "            z=noise\n",
        "            # 6\n",
        "            if Z_DIM == 128:\n",
        "                split_dim = 20\n",
        "                split_dim_remainder = Z_DIM - (split_dim * 5)\n",
        "\n",
        "                z_split = tf.split(z, num_or_size_splits=[split_dim] * 5 + [split_dim_remainder], axis=1)\n",
        "\n",
        "            else:\n",
        "                split_dim = Z_DIM // 6\n",
        "                split_dim_remainder = Z_DIM - (split_dim * 6)\n",
        "\n",
        "                if split_dim_remainder == 0 :\n",
        "                    z_split = tf.split(z, num_or_size_splits=[split_dim] * 6, axis=-1)\n",
        "                else :\n",
        "                    z_split = tf.split(z, num_or_size_splits=[split_dim] * 5 + [split_dim_remainder], axis=1)\n",
        "\n",
        "            ch = G_DIM\n",
        "            #ch=G_DIM\n",
        "            #all_embeddings = lib.get_param(name+'.embed', [NUM_CLASSES, 128], dtype, tf.glorot_uniform_initializer())\n",
        "            #embedding = tf.nn.embedding_lookup(all_embeddings, labels)\n",
        "            print(tf.one_hot(labels, NUM_CLASSES))\n",
        "            if SHARED_EMBED:\n",
        "                embedding = lib.ops.linear.Linear(name+'.embed', NUM_CLASSES, 128, tf.one_hot(labels, NUM_CLASSES), biases=False, spectralnorm=0,weight_regularizer=None)\n",
        "            else:\n",
        "                embedding = tf.one_hot(labels, NUM_CLASSES)\n",
        "            if HIER:    \n",
        "                code=[tf.concat([embedding, z_split[i]],1) for i in range(6)]\n",
        "            else:\n",
        "                code=[embedding]*5\n",
        "            print('code: ', code)\n",
        "            BOTTOM_WIDTH=int(WIDTH//2**np.ceil(np.log2(WIDTH/4)))\n",
        "            NUM_BLOCKS=int(np.log2(WIDTH/BOTTOM_WIDTH))\n",
        "            x = linear(name+'dense',int(z_split[0].shape[-1]), BOTTOM_WIDTH**2 * ch, z_split[0], biases=False)\n",
        "            x = tf.reshape(x, shape=[-1, BOTTOM_WIDTH, BOTTOM_WIDTH, ch])\n",
        "            print('after dense', x)\n",
        "            for k in range(NUM_BLOCKS):\n",
        "                x = GBlock(name+'.up%i'%k, x, code[k+1], channels=ch,is_training=is_training)\n",
        "                print('After Block %i:'%k, x)\n",
        "                #ch = ch // 2\n",
        "                if 2**k*BOTTOM_WIDTH*2==ATTENTION:\n",
        "                    x = self_attention_2(name+'.sa',x, channels=ch)\n",
        "                    print('after sa', x)\n",
        "            \n",
        "            x = batch_norm(name+'.last_bn',x,is_training=is_training)\n",
        "            x = nonlinearity(x)\n",
        "            x = conv2d(name+'.last_conv', int(x.shape[3]), 3, 3, x, stride=1)\n",
        "\n",
        "            x = tf.nn.tanh(x)\n",
        "\n",
        "            return tf.reshape(x, [-1, DATA_DIM])\n",
        "\n",
        "\n",
        "    ##################################################################################\n",
        "    # Discriminator\n",
        "    ##################################################################################\n",
        "\n",
        "def Discriminator(\\\n",
        "    ID, inputs, labels,kp1,kp2,kp3, D_DIM=D_DIM, OUT_DIM=1,\\\n",
        "    code=None, NUM_CHANNELS=3, HEIGHT=HEIGHT, WIDTH=WIDTH, is_training=True):\n",
        "            ch = D_DIM\n",
        "            is_train=True\n",
        "            name='Discriminator.%i'%ID\n",
        "            x = tf.reshape(inputs, [-1, HEIGHT, WIDTH, 3])\n",
        "            print('inputs:', x)\n",
        "            if ADA:\n",
        "              x = augment(x,p=d_p,flip=False)\n",
        "            #x=conv2d(name+'.input', 3, ch, 3, x, stride=1)\n",
        "           # x = nonlinearity(x)\n",
        "        \n",
        "            ch = D_DIM\n",
        "            x = DBlock(name+'.resdown1', x, channels=ch, is_training=is_training,preactivation=False)\n",
        "            x = dropout(x, keep_prob=kp1)\n",
        "            print('after resdown1', x)\n",
        "            \n",
        "            # Non-Local Block\n",
        "            #x = self_attention_2(name+'.sa', x, channels=ch)\n",
        "            #x = dropout(x, keep_prob=kp1)\n",
        "            #print('after sa', x)\n",
        "            \n",
        "            #ch = ch * 2\n",
        "            x = DBlock(name+'.resdown2', x, channels=ch, is_training=is_training)\n",
        "            x = dropout(x, keep_prob=kp1) \n",
        "            print('after resdown2', x)\n",
        "\n",
        "            #ch = ch * 2\n",
        "            #x = DBlock(name+'.resdown4', x, channels=ch, is_training=is_training)\n",
        "            #x = dropout(x, keep_prob=kp1) \n",
        "            #print('after resdown4', x)\n",
        "        \n",
        "            #ch = ch * 2\n",
        "            #x = DBlock(name+'.resdown8', x, channels=ch,  is_training=is_training)\n",
        "           # x = dropout(x, keep_prob=kp1) \n",
        "           # print('after resdown8', x)\n",
        "            #ch = ch * 2\n",
        "            x = DBlock(name+'.res8', x, channels=ch, is_training=is_training, downsample=False)\n",
        "            x = dropout(x, keep_prob=kp2)\n",
        "            print('after res8', x)\n",
        "            #ch = ch * 2\n",
        "            x = DBlock(name+'.res16', x, channels=ch, is_training=is_training, downsample=False)\n",
        "            x = dropout(x, keep_prob=kp2)\n",
        "            print('after res16', x)\n",
        "          #  x = resblock(name+'.reslast', x, channels=ch, is_training=is_training)\n",
        "            x = nonlinearity(x)\n",
        "            #print('after reslast', x)\n",
        "           # x=minibatch_stddev_layer(x)\n",
        "           # x =  lib.ops.conv2d.Conv2D(name+'.refine', int(x.shape[-1]), int(x.shape[-1])-1, 3,x,stride=1)\n",
        "            output2 =  tf.reduce_sum(x, axis=[1,2])\n",
        "\n",
        "         #  shape = output2.get_shape().as_list() # shape is [BS,1,1,n]\n",
        "            #all_embeddings = lib.get_param(name+'.embed', [NUM_CLASSES, shape[1]], dtype, tf.glorot_uniform_initializer())\n",
        "            #embedding = tf.nn.embedding_lookup(all_embeddings, labels)\n",
        "            embedding = linear(name+'.embed', NUM_CLASSES, int(output2.shape[-1]), tf.one_hot(labels, NUM_CLASSES), biases=False, weight_init=tf.random_uniform_initializer(-.1,.1))\n",
        "            output_wgan = linear('Discriminator%i.Output'%ID, ch, 1, output2)\n",
        "            output_wgan = tf.reshape(output_wgan, [-1]) + (tf.reduce_sum(embedding*output2, 1) if CONDITIONAL else 0)    #conrresponding to D\n",
        "            if MH_LOSS or AM_LOSS:\n",
        "                output_mhingegan = linear(name+'.MHINGEGANOutput', D_DIM, NUM_CLASSES+1, output2)\n",
        "                return output_wgan, output2, output_mhingegan\n",
        "            if CONDITIONAL and ACGAN:\n",
        "                output_acgan =linear('Discriminator%i.ACGANOutput'%ID, ch, NUM_CLASSES, output2)\n",
        "                return output_wgan, output2, output_acgan\n",
        "            else:\n",
        "                return output_wgan, output2, None  # two layers' of output\n",
        "\n",
        "\n",
        "    ##################################################################################\n",
        "    # Bottleneck Generator\n",
        "    ##################################################################################\n",
        "\n",
        "def Bottleneck_Generator(ID, noise, n_samples, labels=None):\n",
        "            name='Generator'+'.%i'%ID\n",
        "            z=noise\n",
        "\n",
        "            ch = 16 * G_DIM\n",
        "            embedding = lib.ops.linear.Linear(name+'.embed', NUM_CLASSES, 128, tf.one_hot(labels, NUM_CLASSES))\n",
        "            code=tf.concat([embedding, z],1)\n",
        "            print('code: ', code)\n",
        "            x = linear(name+'dense',int(z.shape[-1]), 4 * 4 * ch, z)\n",
        "            x = tf.reshape(x, shape=[-1, 4, 4, ch])\n",
        "            print('after dense', x)\n",
        "            x=bottleneck_resblock(name+'.16ch.1',x, code, channels=ch, use_bias=False)\n",
        "            x=bottleneck_resblock_up_condition(name+'.16ch.up',x, code, channels=ch, use_bias=False)\n",
        "            x=bottleneck_resblock(name+'.16ch.2',x, code, channels=ch, use_bias=False)\n",
        "            print('after up1', x)  \n",
        "            \n",
        "            ch = ch // 2\n",
        "            x=bottleneck_resblock_up_condition(name+'.8ch.up',x, code, channels=ch, use_bias=False)\n",
        "            x=bottleneck_resblock(name+'.8ch',x, code, channels=ch, use_bias=False)\n",
        "            print('after up2', x)\n",
        "            \n",
        "            ch = ch // 2\n",
        "            x=bottleneck_resblock_up_condition(name+'.4ch.up',x, code, channels=ch, use_bias=False)\n",
        "            x=bottleneck_resblock(name+'.4ch',x, code, channels=ch, use_bias=False)\n",
        "            print('after up3', x)\n",
        "            \n",
        "            ch = ch // 2\n",
        "            x=bottleneck_resblock_up_condition(name+'.2ch.up',x, code, channels=ch, use_bias=False)\n",
        "            print('after up4', x)\n",
        "            # Non-Local Block\n",
        "            x = self_attention_2(name+'.2ch.sa',x, channels=ch)\n",
        "            print('after sa', x)\n",
        "            x=bottleneck_resblock(name+'.2ch',x, code, channels=ch, use_bias=False)\n",
        "\n",
        "            ch = ch // 2\n",
        "            x =bottleneck_resblock_up_condition(name+'.ch.up',x, code, channels=ch, use_bias=False)\n",
        "            print('after up5', x)\n",
        "            x = Normalize('last_bn',x)\n",
        "            x = nonlinearity(x)\n",
        "            x = conv2d(name+'.last_conv', int(x.shape[3]), 3, 3, x, stride=1, biases=False)\n",
        "\n",
        "            x = tf.nn.tanh(x)\n",
        "\n",
        "            return tf.reshape(x, [-1, DATA_DIM])\n",
        "            \n",
        "\n",
        "\n",
        "    ##################################################################################\n",
        "    #Bottleneck Discriminator\n",
        "    ##################################################################################\n",
        "\n",
        "def Bottleneck_Discriminator(\\\n",
        "    ID, inputs, labels,kp1,kp2,kp3, D_DIM=D_DIM, OUT_DIM=1,\\\n",
        "    code=None, NUM_CHANNELS=3, HEIGHT=HEIGHT, WIDTH=WIDTH, is_training=True):\n",
        "            ch = D_DIM\n",
        "            is_train=True\n",
        "            name='Discriminator%i'%ID\n",
        "            x = tf.reshape(inputs, [-1, HEIGHT, WIDTH, 3])\n",
        "            print('inputs:', x)\n",
        "            x=conv2d(name+'.input', 3, ch, 3, x, stride=1, biases=False)\n",
        "            #x = nonlinearity(x)\n",
        "            ch = ch * 2\n",
        "            x = bottleneck_resblock_down(name+'.2ch.down', x, channels=ch, use_bias=False, is_training=is_training)\n",
        "            #x = dropout(x, keep_prob=kp1)\n",
        "            x = bottleneck_resblock(name+'.2ch', x, None, channels=ch, use_bias=False, is_training=is_training)\n",
        "            x = dropout(x, keep_prob=kp1)\n",
        "            print('after resdown1', x)\n",
        "            #x = dropout(x, keep_prob=kp2)\n",
        "            # Non-Local Block\n",
        "            x = self_attention_2(name+'.sa', x, channels=ch)\n",
        "            print('after sa', x)\n",
        "            #x = dropout(x, keep_prob=kp2)\n",
        "            ch = ch * 2\n",
        "            x = bottleneck_resblock_down(name+'.4ch.down', x, channels=ch, use_bias=False, is_training=is_training)\n",
        "            #x = dropout(x, keep_prob=kp2)\n",
        "            x = bottleneck_resblock(name+'.4ch', x, None, channels=ch, use_bias=False, is_training=is_training)\n",
        "            x = dropout(x, keep_prob=kp1)\n",
        "            print('after resdown2', x)    \n",
        "            #x = dropout(x, keep_prob=kp2)\n",
        "            ch = ch * 2\n",
        "            x = bottleneck_resblock_down(name+'.8ch.down', x, channels=ch, use_bias=False, is_training=is_training)\n",
        "            #x = dropout(x, keep_prob=kp2)\n",
        "            x = bottleneck_resblock(name+'.8ch', x, None, channels=ch, use_bias=False, is_training=is_training)\n",
        "            x = dropout(x, keep_prob=kp2)\n",
        "            print('after resdown4', x)\n",
        "            #x = dropout(x, keep_prob=kp2)\n",
        "            ch = ch * 2\n",
        "            x = bottleneck_resblock_down(name+'.16ch.down1', x, channels=ch, use_bias=False, is_training=is_training)\n",
        "            #x = dropout(x, keep_prob=kp2)\n",
        "            x = bottleneck_resblock(name+'.16ch.1', x, None, channels=ch, use_bias=False, is_training=is_training)\n",
        "            x = dropout(x, keep_prob=kp2)\n",
        "            print('after resdown8', x)\n",
        "            #x = dropout(x, keep_prob=kp2)\n",
        "            x = bottleneck_resblock_down(name+'.16ch.down2', x, channels=ch, use_bias=False, is_training=is_training)\n",
        "            #x = dropout(x, keep_prob=kp2)\n",
        "            x = bottleneck_resblock(name+'.16ch.2', x, None, channels=ch, use_bias=False, is_training=is_training)\n",
        "            x = dropout(x, keep_prob=kp2)\n",
        "            print('after resdown16', x)\n",
        "            #x = dropout(x, keep_prob=kp2)\n",
        "            x = nonlinearity(x)\n",
        "            output2 =  tf.reduce_sum(x, axis=[1,2])\n",
        "            embedding = linear(name+'.embed', NUM_CLASSES, int(output2.shape[-1]), tf.one_hot(labels, NUM_CLASSES), weight_init=tf.random_uniform_initializer(-.1,.1))\n",
        "            output_wgan = linear('Discriminator%i.Output'%ID, ch, 1, output2)\n",
        "            output_wgan = tf.reshape(output_wgan, [-1])+tf.reduce_sum(embedding*output2, 1)    #conrresponding to D\n",
        "            if CONDITIONAL and ACGAN:\n",
        "                output_acgan = linear('Discriminator%i.ACGANOutput'%ID, ch, 10, output2)\n",
        "                return output_wgan, output2, output_acgan\n",
        "            else:\n",
        "                return output_wgan, output2, None  # two layers' of output\n",
        "\n",
        "#Generator=Bottleneck_Generator\n",
        "#Discriminator=Bottleneck_Discriminator"
      ],
      "execution_count": 17,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "SvRrryfiZnE_"
      },
      "source": [
        "import tensorflow.compat.v1 as tf\n",
        "import tflib as lib\n",
        "import tflib.custom_ops.custom_ops\n",
        "from typing import Any, List, Tuple, Union\n",
        "IMPL='ref'\n",
        "#IMPL='cuda'\n",
        "from pathlib import Path\n",
        "def _get_plugin(op_name):\n",
        "    cuda_src_dir =os.path.join(os.path.dirname(os.path.realpath('__file__')),'tflib/custom_ops/')\n",
        "    return tflib.custom_ops.custom_ops.get_plugin( cuda_src_dir + op_name + '.cu')\n",
        "class EasyDict(dict):\n",
        "    \"\"\"Convenience class that behaves like a dict but allows access with the attribute syntax.\"\"\"\n",
        "\n",
        "    def __getattr__(self, name: str) -> Any:\n",
        "        try:\n",
        "            return self[name]\n",
        "        except KeyError:\n",
        "            raise AttributeError(name)\n",
        "\n",
        "    def __setattr__(self, name: str, value: Any) -> None:\n",
        "        self[name] = value\n",
        "\n",
        "    def __delattr__(self, name: str) -> None:\n",
        "        del self[name]\n",
        "\n",
        "#----------------------------------------------------------------------------\n",
        "activation_funcs = {\n",
        "    'linear':   EasyDict(func=lambda x, **_:        x,                          def_alpha=None, def_gain=1.0,           cuda_idx=1, ref='y', zero_2nd_grad=True),\n",
        "    'relu':     EasyDict(func=lambda x, **_:        tf.nn.relu(x),              def_alpha=None, def_gain=np.sqrt(2),    cuda_idx=2, ref='y', zero_2nd_grad=True),\n",
        "    'lrelu':    EasyDict(func=lambda x, alpha, **_: tf.nn.leaky_relu(x, alpha), def_alpha=0.2,  def_gain=np.sqrt(2),    cuda_idx=3, ref='y', zero_2nd_grad=True),\n",
        "    'tanh':     EasyDict(func=lambda x, **_:        tf.nn.tanh(x),              def_alpha=None, def_gain=1.0,           cuda_idx=4, ref='y', zero_2nd_grad=False),\n",
        "    'sigmoid':  EasyDict(func=lambda x, **_:        tf.nn.sigmoid(x),           def_alpha=None, def_gain=1.0,           cuda_idx=5, ref='y', zero_2nd_grad=False),\n",
        "    'elu':      EasyDict(func=lambda x, **_:        tf.nn.elu(x),               def_alpha=None, def_gain=1.0,           cuda_idx=6, ref='y', zero_2nd_grad=False),\n",
        "    'selu':     EasyDict(func=lambda x, **_:        tf.nn.selu(x),              def_alpha=None, def_gain=1.0,           cuda_idx=7, ref='y', zero_2nd_grad=False),\n",
        "    'softplus': EasyDict(func=lambda x, **_:        tf.nn.softplus(x),          def_alpha=None, def_gain=1.0,           cuda_idx=8, ref='y', zero_2nd_grad=False),\n",
        "    'swish':    EasyDict(func=lambda x, **_:        tf.nn.sigmoid(x) * x,       def_alpha=None, def_gain=np.sqrt(2),    cuda_idx=9, ref='x', zero_2nd_grad=False),\n",
        "}\n",
        "#----------------------------------------------------------------------------\n",
        "\n",
        "def lerp(a, b, t):\n",
        "    \"\"\"Linear interpolation.\"\"\"\n",
        "    return a + (b - a) * t\n",
        "        \n",
        "def is_tf_expression(x):\n",
        "    \"\"\"Check whether the input is a valid Tensorflow expression, i.e., Tensorflow Tensor, Variable, or Operation.\"\"\"\n",
        "    return isinstance(x, (tf.Tensor, tf.Variable, tf.Operation))\n",
        "    \n",
        "def get_weight(name, shape, gain=1, use_wscale=True, lrmul=1):\n",
        "    fan_in = np.prod(shape[:-1]) # [kernel, kernel, fmaps_in, fmaps_out] or [in, out]\n",
        "    he_std = gain / np.sqrt(fan_in) # He init\n",
        "\n",
        "    # Equalized learning rate and custom learning rate multiplier.\n",
        "    if use_wscale:\n",
        "        init_std = 1.0 / lrmul\n",
        "        runtime_coef = he_std * lrmul\n",
        "    else:\n",
        "        init_std = he_std / lrmul\n",
        "        runtime_coef = lrmul\n",
        "\n",
        "    # Create variable.\n",
        "    init = tf.initializers.random_normal(0, init_std)\n",
        "    return lib.get_param(name, shape=shape, initializer=init) * runtime_coef\n",
        "\n",
        "\n",
        "def upsample_2d(x, k=None, factor=2, gain=1, data_format='NCHW', impl=IMPL):\n",
        "    r\"\"\"Upsample a batch of 2D images with the given filter.\n",
        "\n",
        "    Accepts a batch of 2D images of the shape `[N, C, H, W]` or `[N, H, W, C]`\n",
        "    and upsamples each image with the given filter. The filter is normalized so that\n",
        "    if the input pixels are constant, they will be scaled by the specified `gain`.\n",
        "    Pixels outside the image are assumed to be zero, and the filter is padded with\n",
        "    zeros so that its shape is a multiple of the upsampling factor.\n",
        "\n",
        "    Args:\n",
        "        x:            Input tensor of the shape `[N, C, H, W]` or `[N, H, W, C]`.\n",
        "        k:            FIR filter of the shape `[firH, firW]` or `[firN]` (separable).\n",
        "                      The default is `[1] * factor`, which corresponds to nearest-neighbor\n",
        "                      upsampling.\n",
        "        factor:       Integer upsampling factor (default: 2).\n",
        "        gain:         Scaling factor for signal magnitude (default: 1.0).\n",
        "        data_format:  `'NCHW'` or `'NHWC'` (default: `'NCHW'`).\n",
        "        impl:         Name of the implementation to use. Can be `\"ref\"` or `\"cuda\"` (default).\n",
        "\n",
        "    Returns:\n",
        "        Tensor of the shape `[N, C, H * factor, W * factor]` or\n",
        "        `[N, H * factor, W * factor, C]`, and same datatype as `x`.\n",
        "    \"\"\"\n",
        "\n",
        "    assert isinstance(factor, int) and factor >= 1\n",
        "    if k is None:\n",
        "        k = [1] * factor\n",
        "    k = _setup_kernel(k) * (gain * (factor ** 2))\n",
        "    p = k.shape[0] - factor\n",
        "    return _simple_upfirdn_2d(x, k, up=factor, pad0=(p+1)//2+factor-1, pad1=p//2, data_format=data_format, impl=impl)\n",
        "\n",
        "#----------------------------------------------------------------------------\n",
        "\n",
        "def downsample_2d(x, k=None, factor=2, gain=1, data_format='NCHW', impl=IMPL):\n",
        "    r\"\"\"Downsample a batch of 2D images with the given filter.\n",
        "\n",
        "    Accepts a batch of 2D images of the shape `[N, C, H, W]` or `[N, H, W, C]`\n",
        "    and downsamples each image with the given filter. The filter is normalized so that\n",
        "    if the input pixels are constant, they will be scaled by the specified `gain`.\n",
        "    Pixels outside the image are assumed to be zero, and the filter is padded with\n",
        "    zeros so that its shape is a multiple of the downsampling factor.\n",
        "\n",
        "    Args:\n",
        "        x:            Input tensor of the shape `[N, C, H, W]` or `[N, H, W, C]`.\n",
        "        k:            FIR filter of the shape `[firH, firW]` or `[firN]` (separable).\n",
        "                      The default is `[1] * factor`, which corresponds to average pooling.\n",
        "        factor:       Integer downsampling factor (default: 2).\n",
        "        gain:         Scaling factor for signal magnitude (default: 1.0).\n",
        "        data_format:  `'NCHW'` or `'NHWC'` (default: `'NCHW'`).\n",
        "        impl:         Name of the implementation to use. Can be `\"ref\"` or `\"cuda\"` (default).\n",
        "\n",
        "    Returns:\n",
        "        Tensor of the shape `[N, C, H // factor, W // factor]` or\n",
        "        `[N, H // factor, W // factor, C]`, and same datatype as `x`.\n",
        "    \"\"\"\n",
        "\n",
        "    assert isinstance(factor, int) and factor >= 1\n",
        "    if k is None:\n",
        "        k = [1] * factor\n",
        "    k = _setup_kernel(k) * gain\n",
        "    p = k.shape[0] - factor\n",
        "    return _simple_upfirdn_2d(x, k, down=factor, pad0=(p+1)//2, pad1=p//2, data_format=data_format, impl=impl)\n",
        "\n",
        "#----------------------------------------------------------------------------\n",
        "\n",
        "def upsample_conv_2d(x, w, k=None, factor=2, gain=1, data_format='NCHW', impl=IMPL):\n",
        "    r\"\"\"Fused `upsample_2d()` followed by `tf.nn.conv2d()`.\n",
        "\n",
        "    Padding is performed only once at the beginning, not between the operations.\n",
        "    The fused op is considerably more efficient than performing the same calculation\n",
        "    using standard TensorFlow ops. It supports gradients of arbitrary order.\n",
        "\n",
        "    Args:\n",
        "        x:            Input tensor of the shape `[N, C, H, W]` or `[N, H, W, C]`.\n",
        "        w:            Weight tensor of the shape `[filterH, filterW, inChannels, outChannels]`.\n",
        "                      Grouped convolution can be performed by `inChannels = x.shape[0] // numGroups`.\n",
        "        k:            FIR filter of the shape `[firH, firW]` or `[firN]` (separable).\n",
        "                      The default is `[1] * factor`, which corresponds to nearest-neighbor\n",
        "                      upsampling.\n",
        "        factor:       Integer upsampling factor (default: 2).\n",
        "        gain:         Scaling factor for signal magnitude (default: 1.0).\n",
        "        data_format:  `'NCHW'` or `'NHWC'` (default: `'NCHW'`).\n",
        "        impl:         Name of the implementation to use. Can be `\"ref\"` or `\"cuda\"` (default).\n",
        "\n",
        "    Returns:\n",
        "        Tensor of the shape `[N, C, H * factor, W * factor]` or\n",
        "        `[N, H * factor, W * factor, C]`, and same datatype as `x`.\n",
        "    \"\"\"\n",
        "\n",
        "    assert isinstance(factor, int) and factor >= 1\n",
        "\n",
        "    # Check weight shape.\n",
        "    w = tf.convert_to_tensor(w)\n",
        "    assert w.shape.rank == 4\n",
        "    convH = w.shape[0].value\n",
        "    convW = w.shape[1].value\n",
        "    inC = _shape(w, 2)\n",
        "    outC = _shape(w, 3)\n",
        "    assert convW == convH\n",
        "\n",
        "    # Setup filter kernel.\n",
        "    if k is None:\n",
        "        k = [1] * factor\n",
        "    k = _setup_kernel(k) * (gain * (factor ** 2))\n",
        "    p = (k.shape[0] - factor) - (convW - 1)\n",
        "\n",
        "    # Determine data dimensions.\n",
        "    if data_format == 'NCHW':\n",
        "        stride = [1, 1, factor, factor]\n",
        "        output_shape = [_shape(x, 0), outC, (_shape(x, 2) - 1) * factor + convH, (_shape(x, 3) - 1) * factor + convW]\n",
        "        num_groups = _shape(x, 1) // inC\n",
        "    else:\n",
        "        stride = [1, factor, factor, 1]\n",
        "        output_shape = [_shape(x, 0), (_shape(x, 1) - 1) * factor + convH, (_shape(x, 2) - 1) * factor + convW, outC]\n",
        "        num_groups = _shape(x, 3) // inC\n",
        "\n",
        "    # Transpose weights.\n",
        "    w = tf.reshape(w, [convH, convW, inC, num_groups, -1])\n",
        "    w = tf.transpose(w[::-1, ::-1], [0, 1, 4, 3, 2])\n",
        "    w = tf.reshape(w, [convH, convW, -1, num_groups * inC])\n",
        "\n",
        "    # Execute.\n",
        "    x = tf.nn.conv2d_transpose(x, w, output_shape=output_shape, strides=stride, padding='VALID', data_format=data_format)\n",
        "    return _simple_upfirdn_2d(x, k, pad0=(p+1)//2+factor-1, pad1=p//2+1, data_format=data_format, impl=impl)\n",
        "\n",
        "def _shape(tf_expr, dim_idx):\n",
        "    if tf_expr.shape.rank is not None:\n",
        "        dim = tf_expr.shape[dim_idx].value\n",
        "        if dim is not None:\n",
        "            return dim\n",
        "    return tf.shape(tf_expr)[dim_idx]\n",
        "\n",
        "def _setup_kernel(k):\n",
        "    k = np.asarray(k, dtype=np.float32)\n",
        "    if k.ndim == 1:\n",
        "        k = np.outer(k, k)\n",
        "    k /= np.sum(k)\n",
        "    assert k.ndim == 2\n",
        "    assert k.shape[0] == k.shape[1]\n",
        "    return k\n",
        "\n",
        "\n",
        "def upfirdn_2d(x, k, upx=1, upy=1, downx=1, downy=1, padx0=0, padx1=0, pady0=0, pady1=0, impl=IMPL):\n",
        "    r\"\"\"Pad, upsample, FIR filter, and downsample a batch of 2D images.\n",
        "\n",
        "    Accepts a batch of 2D images of the shape `[majorDim, inH, inW, minorDim]`\n",
        "    and performs the following operations for each image, batched across\n",
        "    `majorDim` and `minorDim`:\n",
        "\n",
        "    1. Pad the image with zeros by the specified number of pixels on each side\n",
        "       (`padx0`, `padx1`, `pady0`, `pady1`). Specifying a negative value\n",
        "       corresponds to cropping the image.\n",
        "\n",
        "    2. Upsample the image by inserting the zeros after each pixel (`upx`, `upy`).\n",
        "\n",
        "    3. Convolve the image with the specified 2D FIR filter (`k`), shrinking the\n",
        "       image so that the footprint of all output pixels lies within the input image.\n",
        "\n",
        "    4. Downsample the image by throwing away pixels (`downx`, `downy`).\n",
        "\n",
        "    This sequence of operations bears close resemblance to scipy.signal.upfirdn().\n",
        "    The fused op is considerably more efficient than performing the same calculation\n",
        "    using standard TensorFlow ops. It supports gradients of arbitrary order.\n",
        "\n",
        "    Args:\n",
        "        x:      Input tensor of the shape `[majorDim, inH, inW, minorDim]`.\n",
        "        k:      2D FIR filter of the shape `[firH, firW]`.\n",
        "        upx:    Integer upsampling factor along the X-axis (default: 1).\n",
        "        upy:    Integer upsampling factor along the Y-axis (default: 1).\n",
        "        downx:  Integer downsampling factor along the X-axis (default: 1).\n",
        "        downy:  Integer downsampling factor along the Y-axis (default: 1).\n",
        "        padx0:  Number of pixels to pad on the left side (default: 0).\n",
        "        padx1:  Number of pixels to pad on the right side (default: 0).\n",
        "        pady0:  Number of pixels to pad on the top side (default: 0).\n",
        "        pady1:  Number of pixels to pad on the bottom side (default: 0).\n",
        "        impl:   Name of the implementation to use. Can be `\"ref\"` or `\"cuda\"` (default).\n",
        "\n",
        "    Returns:\n",
        "        Tensor of the shape `[majorDim, outH, outW, minorDim]`, and same datatype as `x`.\n",
        "    \"\"\"\n",
        "\n",
        "    impl_dict = {\n",
        "        'ref':  _upfirdn_2d_ref,\n",
        "        'cuda': _upfirdn_2d_cuda,\n",
        "    }\n",
        "    return impl_dict[impl](x=x, k=k, upx=upx, upy=upy, downx=downx, downy=downy, padx0=padx0, padx1=padx1, pady0=pady0, pady1=pady1)\n",
        "\n",
        "#----------------------------------------------------------------------------\n",
        "def _upfirdn_2d_cuda(x, k, upx, upy, downx, downy, padx0, padx1, pady0, pady1):\n",
        "    \"\"\"Fast CUDA implementation of `upfirdn_2d()` using custom ops.\"\"\"\n",
        "\n",
        "    x = tf.convert_to_tensor(x)\n",
        "    k = np.asarray(k, dtype=np.float32)\n",
        "    majorDim, inH, inW, minorDim = x.shape.as_list()\n",
        "    kernelH, kernelW = k.shape\n",
        "    assert inW >= 1 and inH >= 1\n",
        "    assert kernelW >= 1 and kernelH >= 1\n",
        "    assert isinstance(upx, int) and isinstance(upy, int)\n",
        "    assert isinstance(downx, int) and isinstance(downy, int)\n",
        "    assert isinstance(padx0, int) and isinstance(padx1, int)\n",
        "    assert isinstance(pady0, int) and isinstance(pady1, int)\n",
        "\n",
        "    outW = (inW * upx + padx0 + padx1 - kernelW) // downx + 1\n",
        "    outH = (inH * upy + pady0 + pady1 - kernelH) // downy + 1\n",
        "    assert outW >= 1 and outH >= 1\n",
        "\n",
        "    kc = tf.constant(k, dtype=x.dtype)\n",
        "    gkc = tf.constant(k[::-1, ::-1], dtype=x.dtype)\n",
        "    gpadx0 = kernelW - padx0 - 1\n",
        "    gpady0 = kernelH - pady0 - 1\n",
        "    gpadx1 = inW * upx - outW * downx + padx0 - upx + 1\n",
        "    gpady1 = inH * upy - outH * downy + pady0 - upy + 1\n",
        "\n",
        "    @tf.custom_gradient\n",
        "    def func(x):\n",
        "        y = _get_plugin('upfirdn_2d').up_fir_dn2d(x=x, k=kc, upx=upx, upy=upy, downx=downx, downy=downy, padx0=padx0, padx1=padx1, pady0=pady0, pady1=pady1)\n",
        "        y.set_shape([majorDim, outH, outW, minorDim])\n",
        "        @tf.custom_gradient\n",
        "        def grad(dy):\n",
        "            dx = _get_plugin('upfirdn_2d').up_fir_dn2d(x=dy, k=gkc, upx=downx, upy=downy, downx=upx, downy=upy, padx0=gpadx0, padx1=gpadx1, pady0=gpady0, pady1=gpady1)\n",
        "            dx.set_shape([majorDim, inH, inW, minorDim])\n",
        "            return dx, func\n",
        "        return y, grad\n",
        "    return func(x)\n",
        "\n",
        "def _upfirdn_2d_ref(x, k, upx, upy, downx, downy, padx0, padx1, pady0, pady1):\n",
        "    \"\"\"Slow reference implementation of `upfirdn_2d()` using standard TensorFlow ops.\"\"\"\n",
        "\n",
        "    x = tf.convert_to_tensor(x)\n",
        "    k = np.asarray(k, dtype=np.float32)\n",
        "    assert x.shape.rank == 4\n",
        "    inH = x.shape[1].value\n",
        "    inW = x.shape[2].value\n",
        "    minorDim = _shape(x, 3)\n",
        "    kernelH, kernelW = k.shape\n",
        "    assert inW >= 1 and inH >= 1\n",
        "    assert kernelW >= 1 and kernelH >= 1\n",
        "    assert isinstance(upx, int) and isinstance(upy, int)\n",
        "    assert isinstance(downx, int) and isinstance(downy, int)\n",
        "    assert isinstance(padx0, int) and isinstance(padx1, int)\n",
        "    assert isinstance(pady0, int) and isinstance(pady1, int)\n",
        "\n",
        "    # Upsample (insert zeros).\n",
        "    x = tf.reshape(x, [-1, inH, 1, inW, 1, minorDim])\n",
        "    x = tf.pad(x, [[0, 0], [0, 0], [0, upy - 1], [0, 0], [0, upx - 1], [0, 0]])\n",
        "    x = tf.reshape(x, [-1, inH * upy, inW * upx, minorDim])\n",
        "\n",
        "    # Pad (crop if negative).\n",
        "    x = tf.pad(x, [[0, 0], [max(pady0, 0), max(pady1, 0)], [max(padx0, 0), max(padx1, 0)], [0, 0]])\n",
        "    x = x[:, max(-pady0, 0) : x.shape[1].value - max(-pady1, 0), max(-padx0, 0) : x.shape[2].value - max(-padx1, 0), :]\n",
        "\n",
        "    # Convolve with filter.\n",
        "    x = tf.transpose(x, [0, 3, 1, 2])\n",
        "    x = tf.reshape(x, [-1, 1, inH * upy + pady0 + pady1, inW * upx + padx0 + padx1])\n",
        "    w = tf.constant(k[::-1, ::-1, np.newaxis, np.newaxis], dtype=x.dtype)\n",
        "    x = tf.nn.conv2d(x, w, strides=[1,1,1,1], padding='VALID', data_format='NCHW')\n",
        "    x = tf.reshape(x, [-1, minorDim, inH * upy + pady0 + pady1 - kernelH + 1, inW * upx + padx0 + padx1 - kernelW + 1])\n",
        "    x = tf.transpose(x, [0, 2, 3, 1])\n",
        "\n",
        "    # Downsample (throw away pixels).\n",
        "    return x[:, ::downy, ::downx, :]\n",
        "\n",
        "def _simple_upfirdn_2d(x, k, up=1, down=1, pad0=0, pad1=0, data_format='NCHW', impl=IMPL):\n",
        "    assert data_format in ['NCHW', 'NHWC']\n",
        "    assert x.shape.rank == 4\n",
        "    y = x\n",
        "    if data_format == 'NCHW':\n",
        "        y = tf.reshape(y, [-1, _shape(y, 2), _shape(y, 3), 1])\n",
        "    y = upfirdn_2d(y, k, upx=up, upy=up, downx=down, downy=down, padx0=pad0, padx1=pad1, pady0=pad0, pady1=pad1, impl=impl)\n",
        "    if data_format == 'NCHW':\n",
        "        y = tf.reshape(y, [-1, _shape(x, 1), _shape(y, 1), _shape(y, 2)])\n",
        "    return y\n",
        "\n",
        "#----------------------------------------------------------------------------\n",
        "\n",
        "def conv_downsample_2d(x, w, k=None, factor=2, gain=1, data_format='NCHW', impl=IMPL):\n",
        "    r\"\"\"Fused `tf.nn.conv2d()` followed by `downsample_2d()`.\n",
        "\n",
        "    Padding is performed only once at the beginning, not between the operations.\n",
        "    The fused op is considerably more efficient than performing the same calculation\n",
        "    using standard TensorFlow ops. It supports gradients of arbitrary order.\n",
        "\n",
        "    Args:\n",
        "        x:            Input tensor of the shape `[N, C, H, W]` or `[N, H, W, C]`.\n",
        "        w:            Weight tensor of the shape `[filterH, filterW, inChannels, outChannels]`.\n",
        "                      Grouped convolution can be performed by `inChannels = x.shape[0] // numGroups`.\n",
        "        k:            FIR filter of the shape `[firH, firW]` or `[firN]` (separable).\n",
        "                      The default is `[1] * factor`, which corresponds to average pooling.\n",
        "        factor:       Integer downsampling factor (default: 2).\n",
        "        gain:         Scaling factor for signal magnitude (default: 1.0).\n",
        "        data_format:  `'NCHW'` or `'NHWC'` (default: `'NCHW'`).\n",
        "        impl:         Name of the implementation to use. Can be `\"ref\"` or `\"cuda\"` (default).\n",
        "\n",
        "    Returns:\n",
        "        Tensor of the shape `[N, C, H // factor, W // factor]` or\n",
        "        `[N, H // factor, W // factor, C]`, and same datatype as `x`.\n",
        "    \"\"\"\n",
        "\n",
        "    assert isinstance(factor, int) and factor >= 1\n",
        "    w = tf.convert_to_tensor(w)\n",
        "    convH, convW, _inC, _outC = w.shape.as_list()\n",
        "    assert convW == convH\n",
        "    if k is None:\n",
        "        k = [1] * factor\n",
        "    k = _setup_kernel(k) * gain\n",
        "    p = (k.shape[0] - factor) + (convW - 1)\n",
        "    if data_format == 'NCHW':\n",
        "        s = [1, 1, factor, factor]\n",
        "    else:\n",
        "        s = [1, factor, factor, 1]\n",
        "    x = _simple_upfirdn_2d(x, k, pad0=(p+1)//2, pad1=p//2, data_format=data_format, impl=impl)\n",
        "    return tf.nn.conv2d(x, w, strides=s, padding='VALID', data_format=data_format)\n",
        "\n",
        "\n",
        "\n",
        "#----------------------------------------------------------------------------\n",
        "# Fully-connected layer.\n",
        "\n",
        "def dense_layer(name, x, fmaps, gain=1, use_wscale=True, lrmul=1):\n",
        "    if len(x.shape) > 2:\n",
        "        x = tf.reshape(x, [-1, np.prod([d.value for d in x.shape[1:]])])\n",
        "    w = get_weight(name+'.Weight', [x.shape[1].value, fmaps], gain=gain, use_wscale=use_wscale, lrmul=lrmul)\n",
        "    w = tf.cast(w, x.dtype)\n",
        "    return tf.matmul(x, w)\n",
        "\n",
        "#----------------------------------------------------------------------------\n",
        "# Convolution layer with optional upsampling or downsampling.\n",
        "\n",
        "def conv2d_layer(name, x, fmaps, kernel, up=False, down=False, resample_kernel=None, gain=1, use_wscale=True, lrmul=1):\n",
        "    assert not (up and down)\n",
        "    assert kernel >= 1 and kernel % 2 == 1\n",
        "    w = get_weight(name+'.Filters', [kernel, kernel, x.shape[1].value, fmaps], gain=gain, use_wscale=use_wscale, lrmul=lrmul)\n",
        "    if up:\n",
        "        x = upsample_conv_2d(x, tf.cast(w, x.dtype), data_format='NCHW', k=resample_kernel)\n",
        "    elif down:\n",
        "        x = conv_downsample_2d(x, tf.cast(w, x.dtype), data_format='NCHW', k=resample_kernel)\n",
        "    else:\n",
        "        x = tf.nn.conv2d(x, tf.cast(w, x.dtype), data_format='NCHW', strides=[1,1,1,1], padding='SAME')\n",
        "    return x\n",
        "\n",
        "\n",
        "#----------------------------------------------------------------------------\n",
        "# Apply bias and activation func.\n",
        "def fused_bias_act(x, b=None, axis=1, act='linear', alpha=None, gain=None, impl=IMPL):\n",
        "    r\"\"\"Fused bias and activation function.\n",
        "\n",
        "    Adds bias `b` to activation tensor `x`, evaluates activation function `act`,\n",
        "    and scales the result by `gain`. Each of the steps is optional. In most cases,\n",
        "    the fused op is considerably more efficient than performing the same calculation\n",
        "    using standard TensorFlow ops. It supports first and second order gradients,\n",
        "    but not third order gradients.\n",
        "\n",
        "    Args:\n",
        "        x:      Input activation tensor. Can have any shape, but if `b` is defined, the\n",
        "                dimension corresponding to `axis`, as well as the rank, must be known.\n",
        "        b:      Bias vector, or `None` to disable. Must be a 1D tensor of the same type\n",
        "                as `x`. The shape must be known, and it must match the dimension of `x`\n",
        "                corresponding to `axis`.\n",
        "        axis:   The dimension in `x` corresponding to the elements of `b`.\n",
        "                The value of `axis` is ignored if `b` is not specified.\n",
        "        act:    Name of the activation function to evaluate, or `\"linear\"` to disable.\n",
        "                Can be e.g. `\"relu\"`, `\"lrelu\"`, `\"tanh\"`, `\"sigmoid\"`, `\"swish\"`, etc.\n",
        "                See `activation_funcs` for a full list. `None` is not allowed.\n",
        "        alpha:  Shape parameter for the activation function, or `None` to use the default.\n",
        "        gain:   Scaling factor for the output tensor, or `None` to use default.\n",
        "                See `activation_funcs` for the default scaling of each activation function.\n",
        "                If unsure, consider specifying `1.0`.\n",
        "        impl:   Name of the implementation to use. Can be `\"ref\"` or `\"cuda\"` (default).\n",
        "\n",
        "    Returns:\n",
        "        Tensor of the same shape and datatype as `x`.\n",
        "    \"\"\"\n",
        "\n",
        "    impl_dict = {\n",
        "        'ref':  _fused_bias_act_ref,\n",
        "        'cuda': _fused_bias_act_cuda,\n",
        "    }\n",
        "    return impl_dict[impl](x=x, b=b, axis=axis, act=act, alpha=alpha, gain=gain)\n",
        "\n",
        "#----------------------------------------------------------------------------\n",
        "def _fused_bias_act_cuda(x, b, axis, act, alpha, gain):\n",
        "    \"\"\"Fast CUDA implementation of `fused_bias_act()` using custom ops.\"\"\"\n",
        "\n",
        "    # Validate arguments.\n",
        "    x = tf.convert_to_tensor(x)\n",
        "    empty_tensor = tf.constant([], dtype=x.dtype)\n",
        "    b = tf.convert_to_tensor(b) if b is not None else empty_tensor\n",
        "    act_spec = activation_funcs[act]\n",
        "    assert b.shape.rank == 1 and (b.shape[0] == 0 or b.shape[0] == x.shape[axis])\n",
        "    assert b.shape[0] == 0 or 0 <= axis < x.shape.rank\n",
        "    if alpha is None:\n",
        "        alpha = act_spec.def_alpha\n",
        "    if gain is None:\n",
        "        gain = act_spec.def_gain\n",
        "\n",
        "    # Special cases.\n",
        "    if act == 'linear' and b is None and gain == 1.0:\n",
        "        return x\n",
        "    if act_spec.cuda_idx is None:\n",
        "        return _fused_bias_act_ref(x=x, b=b, axis=axis, act=act, alpha=alpha, gain=gain)\n",
        "\n",
        "    # CUDA kernel.\n",
        "    cuda_kernel = _get_plugin('fused_bias_act').fused_bias_act\n",
        "    cuda_kwargs = dict(axis=axis, act=act_spec.cuda_idx, alpha=alpha, gain=gain)\n",
        "\n",
        "    # Forward pass: y = func(x, b).\n",
        "    def func_y(x, b):\n",
        "        y = cuda_kernel(x=x, b=b, ref=empty_tensor, grad=0, **cuda_kwargs)\n",
        "        y.set_shape(x.shape)\n",
        "        return y\n",
        "\n",
        "    # Backward pass: dx, db = grad(dy, x, y)\n",
        "    def grad_dx(dy, x, y):\n",
        "        ref = {'x': x, 'y': y}[act_spec.ref]\n",
        "        dx = cuda_kernel(x=dy, b=empty_tensor, ref=ref, grad=1, **cuda_kwargs)\n",
        "        dx.set_shape(x.shape)\n",
        "        return dx\n",
        "    def grad_db(dx):\n",
        "        if b.shape[0] == 0:\n",
        "            return empty_tensor\n",
        "        db = dx\n",
        "        if axis < x.shape.rank - 1:\n",
        "            db = tf.reduce_sum(db, list(range(axis + 1, x.shape.rank)))\n",
        "        if axis > 0:\n",
        "            db = tf.reduce_sum(db, list(range(axis)))\n",
        "        db.set_shape(b.shape)\n",
        "        return db\n",
        "\n",
        "    # Second order gradients: d_dy, d_x = grad2(d_dx, d_db, x, y)\n",
        "    def grad2_d_dy(d_dx, d_db, x, y):\n",
        "        ref = {'x': x, 'y': y}[act_spec.ref]\n",
        "        d_dy = cuda_kernel(x=d_dx, b=d_db, ref=ref, grad=1, **cuda_kwargs)\n",
        "        d_dy.set_shape(x.shape)\n",
        "        return d_dy\n",
        "    def grad2_d_x(d_dx, d_db, x, y):\n",
        "        ref = {'x': x, 'y': y}[act_spec.ref]\n",
        "        d_x = cuda_kernel(x=d_dx, b=d_db, ref=ref, grad=2, **cuda_kwargs)\n",
        "        d_x.set_shape(x.shape)\n",
        "        return d_x\n",
        "\n",
        "    # Fast version for piecewise-linear activation funcs.\n",
        "    @tf.custom_gradient\n",
        "    def func_zero_2nd_grad(x, b):\n",
        "        y = func_y(x, b)\n",
        "        @tf.custom_gradient\n",
        "        def grad(dy):\n",
        "            dx = grad_dx(dy, x, y)\n",
        "            db = grad_db(dx)\n",
        "            def grad2(d_dx, d_db):\n",
        "                d_dy = grad2_d_dy(d_dx, d_db, x, y)\n",
        "                return d_dy\n",
        "            return (dx, db), grad2\n",
        "        return y, grad\n",
        "\n",
        "    # Slow version for general activation funcs.\n",
        "    @tf.custom_gradient\n",
        "    def func_nonzero_2nd_grad(x, b):\n",
        "        y = func_y(x, b)\n",
        "        def grad_wrap(dy):\n",
        "            @tf.custom_gradient\n",
        "            def grad_impl(dy, x):\n",
        "                dx = grad_dx(dy, x, y)\n",
        "                db = grad_db(dx)\n",
        "                def grad2(d_dx, d_db):\n",
        "                    d_dy = grad2_d_dy(d_dx, d_db, x, y)\n",
        "                    d_x = grad2_d_x(d_dx, d_db, x, y)\n",
        "                    return d_dy, d_x\n",
        "                return (dx, db), grad2\n",
        "            return grad_impl(dy, x)\n",
        "        return y, grad_wrap\n",
        "\n",
        "    # Which version to use?\n",
        "    if act_spec.zero_2nd_grad:\n",
        "        return func_zero_2nd_grad(x, b)\n",
        "    return func_nonzero_2nd_grad(x, b)\n",
        "\n",
        "#----------------------------------------------------------------------------\n",
        "\n",
        "def _fused_bias_act_ref(x, b, axis, act, alpha, gain):\n",
        "    \"\"\"Slow reference implementation of `fused_bias_act()` using standard TensorFlow ops.\"\"\"\n",
        "\n",
        "    # Validate arguments.\n",
        "    x = tf.convert_to_tensor(x)\n",
        "    b = tf.convert_to_tensor(b) if b is not None else tf.constant([], dtype=x.dtype)\n",
        "    act_spec = activation_funcs[act]\n",
        "    assert b.shape.rank == 1 and (b.shape[0] == 0 or b.shape[0] == x.shape[axis])\n",
        "    assert b.shape[0] == 0 or 0 <= axis < x.shape.rank\n",
        "    if alpha is None:\n",
        "        alpha = act_spec.def_alpha\n",
        "    if gain is None:\n",
        "        gain = act_spec.def_gain\n",
        "\n",
        "    # Add bias.\n",
        "    if b.shape[0] != 0:\n",
        "        x += tf.reshape(b, [-1 if i == axis else 1 for i in range(x.shape.rank)])\n",
        "\n",
        "    # Evaluate activation function.\n",
        "    x = act_spec.func(x, alpha=alpha)\n",
        "\n",
        "    # Scale by gain.\n",
        "    if gain != 1:\n",
        "        x *= gain\n",
        "    return x\n",
        "\n",
        "def apply_bias_act(name, x, act='linear', alpha=None, gain=None, lrmul=1):\n",
        "    b = lib.get_param(name+'.Bias', shape=[x.shape[1]], initializer=tf.initializers.zeros()) * lrmul\n",
        "    return fused_bias_act(x, b=tf.cast(b, x.dtype), act=act, alpha=alpha, gain=gain)\n",
        "\n",
        "#----------------------------------------------------------------------------\n",
        "# Naive upsampling (nearest neighbor) and downsampling (average pooling).\n",
        "\n",
        "def naive_upsample_2d(x, factor=2):\n",
        "        _N, C, H, W = x.shape.as_list()\n",
        "        x = tf.reshape(x, [-1, C, H, 1, W, 1])\n",
        "        x = tf.tile(x, [1, 1, 1, factor, 1, factor])\n",
        "        return tf.reshape(x, [-1, C, H * factor, W * factor])\n",
        "\n",
        "def naive_downsample_2d(x, factor=2):\n",
        "        _N, C, H, W = x.shape.as_list()\n",
        "        x = tf.reshape(x, [-1, C, H // factor, factor, W // factor, factor])\n",
        "        return tf.reduce_mean(x, axis=[3,5])\n",
        "\n",
        "#----------------------------------------------------------------------------\n",
        "# Modulated convolution layer.\n",
        "\n",
        "def modulated_conv2d_layer(name, x, y, fmaps, kernel, up=False, down=False, demodulate=True, resample_kernel=None, gain=1, use_wscale=True, lrmul=1, fused_modconv=True):\n",
        "    assert not (up and down)\n",
        "    assert kernel >= 1 and kernel % 2 == 1\n",
        "\n",
        "    # Get weight.\n",
        "    w = get_weight(name+'.weight', [kernel, kernel, x.shape[1].value, fmaps], gain=gain, use_wscale=use_wscale, lrmul=lrmul)\n",
        "    ww = w[np.newaxis] # [BkkIO] Introduce minibatch dimension.\n",
        "\n",
        "    # Modulate.\n",
        "    s = dense_layer(name+'.mod_weight', y, fmaps=x.shape[1].value) # [BI] Transform incoming W to style.\n",
        "    s = apply_bias_act(name+'.mod_bias', s) + 1 # [BI] Add bias (initially 1).\n",
        "    ww *= tf.cast(s[:, np.newaxis, np.newaxis, :, np.newaxis], w.dtype) # [BkkIO] Scale input feature maps.\n",
        "\n",
        "    # Demodulate.\n",
        "    if demodulate:\n",
        "        d = tf.rsqrt(tf.reduce_sum(tf.square(ww), axis=[1,2,3]) + 1e-8) # [BO] Scaling factor.\n",
        "        ww *= d[:, np.newaxis, np.newaxis, np.newaxis, :] # [BkkIO] Scale output feature maps.\n",
        "\n",
        "    # Reshape/scale input.\n",
        "    if fused_modconv:\n",
        "        x = tf.reshape(x, [1, -1, x.shape[2], x.shape[3]]) # Fused => reshape minibatch to convolution groups.\n",
        "        w = tf.reshape(tf.transpose(ww, [1, 2, 3, 0, 4]), [ww.shape[1], ww.shape[2], ww.shape[3], -1])\n",
        "    else:\n",
        "        x *= tf.cast(s[:, :, np.newaxis, np.newaxis], x.dtype) # [BIhw] Not fused => scale input activations.\n",
        "\n",
        "    # Convolution with optional up/downsampling.\n",
        "    if up:\n",
        "        x = upsample_conv_2d(x, tf.cast(w, x.dtype), data_format='NCHW', k=resample_kernel)\n",
        "    elif down:\n",
        "        x = conv_downsample_2d(x, tf.cast(w, x.dtype), data_format='NCHW', k=resample_kernel)\n",
        "    else:\n",
        "        x = tf.nn.conv2d(x, tf.cast(w, x.dtype), data_format='NCHW', strides=[1,1,1,1], padding='SAME')\n",
        "\n",
        "    # Reshape/scale output.\n",
        "    if fused_modconv:\n",
        "        x = tf.reshape(x, [-1, fmaps, x.shape[2], x.shape[3]]) # Fused => reshape convolution groups back to minibatch.\n",
        "    elif demodulate:\n",
        "        x *= tf.cast(d[:, :, np.newaxis, np.newaxis], x.dtype) # [BOhw] Not fused => scale output activations.\n",
        "    return x\n",
        "\n",
        "#----------------------------------------------------------------------------\n",
        "# Minibatch standard deviation layer.\n",
        "def minibatch_stddev_layer(x, group_size=None, num_new_features=1):\n",
        "    if group_size is None:\n",
        "        group_size = tf.shape(x)[0]\n",
        "    else:\n",
        "        group_size = tf.minimum(group_size, tf.shape(x)[0]) # Minibatch must be divisible by (or smaller than) group_size.\n",
        "    G = group_size\n",
        "    F = num_new_features\n",
        "    _N, C, H, W = x.shape.as_list()\n",
        "    c = C // F\n",
        "    y = tf.cast(x, tf.float32)                # [NCHW]   Cast to FP32.\n",
        "    y = tf.reshape(y, [G, -1, F, c, H, W])    # [GnFcHW] Split minibatch N into n groups of size G, and channels C into F groups of size c.\n",
        "    y -= tf.reduce_mean(y, axis=0)            # [GnFcHW] Subtract mean over group.\n",
        "    y = tf.reduce_mean(tf.square(y), axis=0)  # [nFcHW]  Calc variance over group.\n",
        "    y = tf.sqrt(y + 1e-8)                     # [nFcHW]  Calc stddev over group.\n",
        "    y = tf.reduce_mean(y, axis=[2,3,4])       # [nF]     Take average over channels and pixels.\n",
        "    y = tf.cast(y, x.dtype)                   # [nF]     Cast back to original data type.\n",
        "    y = tf.reshape(y, [-1, F, 1, 1])          # [nF11]   Add missing dimensions.\n",
        "    y = tf.tile(y, [G, 1, H, W])              # [NFHW]   Replicate over group and pixels.\n",
        "    return tf.concat([x, y], axis=1)          # [NCHW]   Append to input as new channels.\n",
        "\n",
        "def normalize_2nd_moment(x, axis=1, eps=1e-8):\n",
        "    return x * tf.rsqrt(tf.reduce_mean(tf.square(x), axis=axis, keepdims=True) + eps)     \n",
        "def G_mapping(\n",
        "    name,\n",
        "    latents_in,                             # First input: Latent vectors (Z) [minibatch, latent_size].\n",
        "    labels_in,                              # Second input: Conditioning labels [minibatch, label_size].\n",
        "    latent_size             = 512,          # Latent vector (Z) dimensionality.\n",
        "    label_size              = NUM_CLASSES,            # Label dimensionality, 0 if no labels.\n",
        "    dlatent_size            = 512,          # Disentangled latent (W) dimensionality.\n",
        "    dlatent_broadcast       = None,         # Output disentangled latent (W) as [minibatch, dlatent_size] or [minibatch, dlatent_broadcast, dlatent_size].\n",
        "    mapping_layers          = 2,            # Number of mapping layers.\n",
        "    mapping_fmaps           = 512,          # Number of activations in the mapping layers.\n",
        "    mapping_lrmul           = 0.01,         # Learning rate multiplier for the mapping layers.\n",
        "    mapping_nonlinearity    = 'lrelu',      # Activation function: 'relu', 'lrelu', etc.\n",
        "    normalize_latents       = True,         # Normalize latent vectors (Z) before feeding them to the mapping layers?\n",
        "    dtype                   = 'float32',    # Data type to use for activations and outputs.\n",
        "    **_kwargs):                             # Ignore unrecognized keyword args.\n",
        "\n",
        "    act = mapping_nonlinearity\n",
        "\n",
        "    # Inputs.\n",
        "    latents_in.set_shape([None, latent_size])\n",
        "    labels_in.set_shape([None, label_size])\n",
        "    latents_in = tf.cast(latents_in, dtype)\n",
        "    labels_in = tf.cast(labels_in, dtype)\n",
        "    x = latents_in\n",
        "\n",
        "    # Normalize latents.\n",
        "    if normalize_latents:\n",
        "           x = normalize_2nd_moment(x)\n",
        "    # Embed labels and concatenate them with latents.\n",
        "    if label_size:\n",
        "            y = labels_in\n",
        "            y = apply_bias_act(name+'.embed.bias',dense_layer(name+'.embed.dense', y, fmaps=dlatent_size))\n",
        "            y = normalize_2nd_moment(y)\n",
        "            x = tf.concat([x, y], axis=1)\n",
        "\n",
        "    # Mapping layers.\n",
        "    for layer_idx in range(mapping_layers):\n",
        "            fmaps = dlatent_size if layer_idx == mapping_layers - 1 else mapping_fmaps\n",
        "            x = apply_bias_act(name+'.Bias_act%d' % layer_idx, dense_layer(name+'.Dense%d' % layer_idx, x, fmaps=fmaps, lrmul=mapping_lrmul), act=act, lrmul=mapping_lrmul)\n",
        "\n",
        "    # Broadcast.\n",
        "    if dlatent_broadcast is not None:\n",
        "            x = tf.tile(x[:, np.newaxis], [1, dlatent_broadcast, 1])\n",
        "\n",
        "    # Output.\n",
        "    assert x.dtype == tf.as_dtype(dtype)\n",
        "    return x\n",
        "\n",
        "#----------------------------------------------------------------------------\n",
        "# StyleGAN synthesis network with revised architecture (Figure 2d).\n",
        "# Implements progressive growing, but no skip connections or residual nets (Figure 7).\n",
        "# Used in configs B-D (Table 1).\n",
        "def random_normal(name, shape, dtype=tf.float32, update_noise=False):\n",
        "    noise=lib.get_param(name, shape=shape, dtype=dtype, initializer=tf.initializers.random_normal(), trainable=False)\n",
        "    if update_noise:\n",
        "        with tf.control_dependencies([noise.initializer]):\n",
        "            return tf.identity(noise)\n",
        "    return noise\n",
        "    \n",
        "def random_uniform(name, shape, minval, maxval, dtype=tf.float32, update_noise=False):\n",
        "    noise=lib.get_param(name, shape=shape, dtype=dtype, initializer=tf.initializers.random_uniform(minval=minval, maxval=maxval, dtype=dtype), trainable=False)\n",
        "    if update_noise:\n",
        "        with tf.control_dependencies([noise.initializer]):\n",
        "            return tf.identity(noise)\n",
        "    return noise\n",
        "def G_synthesis_stylegan2(\n",
        "    name,\n",
        "    latents_in,\n",
        "    dlatents_in,                        # Input: Disentangled latents (W) [minibatch, num_layers, dlatent_size].\n",
        "    dlatent_size        = 512,          # Disentangled latent (W) dimensionality.\n",
        "    num_channels        = 3,            # Number of output color channels.\n",
        "    resolution          = 32,         # Output resolution.\n",
        "    fmap_base           = 16 << 10,     # Overall multiplier for the number of feature maps.\n",
        "    fmap_decay          = 1.0,          # log2 feature map reduction when doubling the resolution.\n",
        "    fmap_min            = 1,            # Minimum number of feature maps in any layer.\n",
        "    fmap_max            = 512,          # Maximum number of feature maps in any layer.\n",
        "    randomize_noise     = True,         # True = randomize noise inputs every time (non-deterministic), False = read noise inputs from variables.\n",
        "    architecture        = 'skip',       # Architecture: 'orig', 'skip', 'resnet'.\n",
        "    nonlinearity        = 'lrelu',      # Activation function: 'relu', 'lrelu', etc.\n",
        "    dtype               = dtype,    # Data type to use for activations and outputs.\n",
        "    resample_kernel     = [1,3,3,1],    # Low-pass filter to apply when resampling activations. None = no filtering.\n",
        "    fused_modconv       = False,         # Implement modulated_conv2d_layer() as a single fused op?\n",
        "    update_noise=True,\n",
        "    **_kwargs):                         # Ignore unrecognized keyword args.\n",
        "\n",
        "    resolution_log2 = int(np.log2(resolution))\n",
        "    assert resolution == 2**resolution_log2 and resolution >= 4\n",
        "    def nf(stage): return np.clip(int(fmap_base / (2.0 ** (stage * fmap_decay))), fmap_min, fmap_max)\n",
        "    assert architecture in ['orig', 'skip', 'resnet']\n",
        "    act = nonlinearity\n",
        "    num_layers = resolution_log2 * 2 - 2\n",
        "    images_out = None\n",
        "\n",
        "    # Primary inputs.\n",
        "    dlatents_in.set_shape([None, num_layers, dlatent_size])\n",
        "    dlatents_in = tf.cast(dlatents_in, dtype)\n",
        "\n",
        "    # Noise inputs.\n",
        "    noise_inputs = []\n",
        "    reused_noise_inputs =[]\n",
        "    for layer_idx in range(num_layers - 1):\n",
        "        res = (layer_idx + 5) // 2\n",
        "        shape = [1, 1, 2**res, 2**res]\n",
        "        noise_inputs.append(lib.get_param(name+'.noise%d' % layer_idx, shape=shape, initializer=tf.initializers.random_normal(), trainable=False))\n",
        "    \n",
        "    #Reuse the input latent noise as the addictive noise\n",
        "    #latents_in=tf.concat([latents_in,tf.reshape(tf.transpose(latents_in),latents_in.shape)],0)\n",
        "    #reuesed_latents_in=[tf.split(tf.reshape(tf.roll(latents_in,-idx-1,0),[-1]),[4**2,8**2,8**2,16**2,16**2,32**2,32**2,-1]) for idx in range(BATCH_SIZE)]\n",
        "    #for layer_idx in range(num_layers - 1):\n",
        "        #res = (layer_idx + 5) // 2\n",
        "       # shape = [BATCH_SIZE, 1, 2**res, 2**res]\n",
        "        #reused_noise_inputs.append(tf.reshape([reuesed_latents_in[idx][layer_idx] for idx in range(BATCH_SIZE)], shape))\n",
        "    # Single convolution layer with all the bells and whistles.\n",
        "    def layer(name, x, layer_idx, fmaps, kernel, up=False):\n",
        "        x = modulated_conv2d_layer(name+'.Mod_conv', x, dlatents_in[:, layer_idx], fmaps=fmaps, kernel=kernel, up=up, resample_kernel=resample_kernel, fused_modconv=fused_modconv)\n",
        "        if randomize_noise:\n",
        "            #noise = random_normal(name+'layer_noise', [8, x.shape[0], 1, x.shape[2], x.shape[3]], dtype=x.dtype, update_noise=update_noise)\n",
        "            #replica_id = tf.distribute.get_replica_context().replica_id_in_sync_group\n",
        "            #noise = noise[replica_id]\n",
        "            noise = tf.random_normal([tf.shape(x)[0], 1, x.shape[2], x.shape[3]], dtype=x.dtype)\n",
        "        else:\n",
        "            noise = tf.cast(noise_inputs[layer_idx], x.dtype)\n",
        "        noise_strength = lib.get_param(name+'.noise_strength', shape=[], initializer=tf.initializers.zeros())\n",
        "        x += noise * tf.cast(noise_strength, x.dtype)\n",
        "        return apply_bias_act(name+'.Bias', x, act=act)\n",
        "\n",
        "    # Building blocks for main layers.\n",
        "    def block(name, x, res): # res = 3..resolution_log2\n",
        "        t = x\n",
        "        x = layer(name+'.%dx%d.Conv0_up' % (2**res, 2**res), x, layer_idx=res*2-5, fmaps=nf(res-1), kernel=3, up=True)\n",
        "        x = layer(name+'.%dx%d.Conv1' % (2**res, 2**res), x, layer_idx=res*2-4, fmaps=nf(res-1), kernel=3)\n",
        "        if architecture == 'resnet':\n",
        "                t = conv2d_layer(name+'.%dx%d.Skip' % (2**res, 2**res), t, fmaps=nf(res-1), kernel=1, up=True, resample_kernel=resample_kernel)\n",
        "                x = (x + t) * (1 / np.sqrt(2))\n",
        "        return x\n",
        "    def upsample(y):\n",
        "            return upsample_2d(y, k=resample_kernel)\n",
        "    def torgb(name, x, y, res): # res = 2..resolution_log2\n",
        "        name=name+'.ToRGB_lod%d' % (resolution_log2 - res)\n",
        "        t = apply_bias_act(name+'.Bias', modulated_conv2d_layer(name+'.Mod_conv', x, dlatents_in[:, res*2-3], fmaps=num_channels, kernel=1, demodulate=False, fused_modconv=fused_modconv))\n",
        "        return t if y is None else y + t\n",
        "\n",
        "    # Early layers.\n",
        "    y = None\n",
        "    x = lib.get_param(name+'.Const4x4', shape=[1, nf(1), 4, 4], initializer=tf.initializers.random_normal())\n",
        "    x = tf.tile(tf.cast(x, dtype), [tf.shape(dlatents_in)[0], 1, 1, 1])\n",
        "    x = layer(name+'.Conv4x4', x, layer_idx=0, fmaps=nf(1), kernel=3)\n",
        "    if architecture == 'skip':\n",
        "        y = torgb(name+'.ToRGB4x4', x, y, 2)\n",
        "\n",
        "    # Main layers.\n",
        "    for res in range(3, resolution_log2 + 1):\n",
        "            x = block(name+'.B%dx%d' % (2**res, 2**res), x, res)\n",
        "            print('after GBlock%dx%d:'%(2**res, 2**res), x)\n",
        "            if architecture == 'skip':\n",
        "                y = upsample(y)\n",
        "            if architecture == 'skip' or res == resolution_log2:\n",
        "                y = torgb(name+'.ToRGB%dx%d' % (2**res, 2**res), x, y, res)\n",
        "    images_out = y\n",
        "\n",
        "    assert images_out.dtype == tf.as_dtype(dtype)\n",
        "    return images_out\n",
        "    \n",
        "\n",
        "            \n",
        "def G_main(\n",
        "    ID,\n",
        "    noise,                                         # First input: Latent vectors (Z) [minibatch, latent_size].\n",
        "    n_samples,\n",
        "    labels,                                          # Second input: Conditioning labels [minibatch, label_size].\n",
        "    NUM_CLASSES=10,\n",
        "    ema=False,\n",
        "    dlatent_size=512,\n",
        "    resolution=32,\n",
        "    truncation_psi          = 0.5,                      # Style strength multiplier for the truncation trick. None = disable.\n",
        "    truncation_cutoff       = None,                     # Number of layers for which to apply the truncation trick. None = disable.\n",
        "    truncation_psi_val      = None,                     # Value for truncation_psi to use during validation.\n",
        "    truncation_cutoff_val   = None,                     # Value for truncation_cutoff to use during validation.\n",
        "    dlatent_avg_beta        = 0.995,                    # Decay for tracking the moving average of W during training. None = disable.\n",
        "    style_mixing_prob       = None,                      # Probability of mixing styles during training. None = disable.\n",
        "    is_training             = True,                    # Network is under training? Enables and disables specific features.\n",
        "    is_validation           = False,                    # Network is under validation? Chooses which value to use for truncation_psi.\n",
        "    return_dlatents         = False,                    # Return dlatents in addition to the images?\n",
        "    is_template_graph       = False,                    # True = template graph constructed by the Network class, False = actual evaluation.\n",
        "    synthesis_func          = 'G_synthesis_stylegan2',  # Build func name for the synthesis network.\n",
        "    update_sn=None,\n",
        "    update_noise=True,\n",
        "    **kwargs):                                          # Arguments for sub-networks (mapping and synthesis).\n",
        "    name= 'Generator%i_'%ID if not ema else 'Generator%i_ema_'%ID\n",
        "    is_validation= not is_training\n",
        "    latents_in=noise\n",
        "    labels_in=tf.one_hot(labels,NUM_CLASSES)\n",
        "    resolution_log2 = int(np.log2(resolution))\n",
        "    num_layers = resolution_log2 * 2 - 2\n",
        "    \n",
        "    # Validate arguments.\n",
        "    assert not is_training or not is_validation\n",
        "    if is_validation:\n",
        "        truncation_psi = truncation_psi_val\n",
        "        truncation_cutoff = truncation_cutoff_val\n",
        "    if is_training or (truncation_psi is not None and not is_tf_expression(truncation_psi) and truncation_psi == 1):\n",
        "        truncation_psi = None\n",
        "    if is_training:\n",
        "        truncation_cutoff = None\n",
        "    if not is_training or (dlatent_avg_beta is not None and not is_tf_expression(dlatent_avg_beta) and dlatent_avg_beta == 1):\n",
        "        dlatent_avg_beta = None\n",
        "    if not is_training or (style_mixing_prob is not None and not is_tf_expression(style_mixing_prob) and style_mixing_prob <= 0):\n",
        "        style_mixing_prob = None\n",
        "        \n",
        "\n",
        "\n",
        "    # Setup variables.\n",
        "    dlatent_avg = lib.get_param(name+'.dlatent_avg', shape=[dlatent_size], initializer=tf.initializers.zeros(), trainable=False)\n",
        "\n",
        "    # Evaluate mapping network.\n",
        "    dlatents = G_mapping(name+'.mapping', latents_in, labels_in, is_training=is_training, dlatent_broadcast=num_layers)\n",
        "    dlatents = tf.cast(dlatents, tf.float32)\n",
        "\n",
        "\n",
        "    # Update moving average of W.\n",
        "    if dlatent_avg_beta is not None:\n",
        "            #batch_avg = cross_replica_mean(dlatents[:, 0], axis=0)\n",
        "            batch_avg = tf.reduce_mean(dlatents[:, 0], axis=0)\n",
        "            update_op = tf.assign(dlatent_avg, lerp(batch_avg, dlatent_avg, dlatent_avg_beta))\n",
        "            with tf.control_dependencies([update_op]):\n",
        "                dlatents = tf.identity(dlatents)\n",
        "\n",
        "    # Perform style mixing regularization.\n",
        "    if style_mixing_prob is not None:\n",
        "            #latents2 = tf.roll(latents_in, 1, 0)\n",
        "            latents2 = tf.random_normal(tf.shape(latents_in))\n",
        "            dlatents2 = G_mapping(name+'.mapping', latents2, labels_in, is_training=is_training, dlatent_broadcast=num_layers)\n",
        "            dlatents2 = tf.cast(dlatents2, tf.float32)\n",
        "            layer_idx = np.arange(num_layers)[np.newaxis, :, np.newaxis]\n",
        "            cur_layers = num_layers# - tf.cast(lod_in, tf.int32) * 2\n",
        "            cutoff_layer=random_uniform(name+'.cutoff_layer', shape=[], minval=1, maxval=cur_layers, dtype=tf.int32, update_noise=update_noise)\n",
        "            use_style_mixing=random_uniform(name+'.use_style_mixing', shape=[], minval=0, maxval=1, dtype=tf.float32, update_noise=update_noise)\n",
        "            mixing_cutoff = tf.cond(\n",
        "                use_style_mixing < style_mixing_prob,\n",
        "                lambda: cutoff_layer,\n",
        "                lambda: cur_layers)        \n",
        "            dlatents = tf.where(tf.broadcast_to(layer_idx < mixing_cutoff, tf.shape(dlatents)), dlatents, dlatents2)\n",
        "\n",
        "    # Apply truncation trick.\n",
        "    if truncation_psi is not None:\n",
        "            layer_idx = np.arange(num_layers)[np.newaxis, :, np.newaxis]\n",
        "            layer_psi = np.ones(layer_idx.shape, dtype=np.float32)\n",
        "            if truncation_cutoff is None:\n",
        "                layer_psi *= truncation_psi\n",
        "            else:\n",
        "                layer_psi = tf.where(layer_idx < truncation_cutoff, layer_psi * truncation_psi, layer_psi)\n",
        "            dlatents = lerp(dlatent_avg, dlatents, layer_psi)\n",
        "\n",
        "    # Evaluate synthesis network.\n",
        "    deps = []\n",
        "\n",
        "    with tf.control_dependencies(deps):\n",
        "        synth_net= G_synthesis_stylegan2 if synthesis_func == 'G_synthesis_stylegan2' else G_synthesis_stylegan_revised\n",
        "        images_out = synth_net(name+'.Synthesis', latents_in, dlatents, is_training=is_training, force_clean_graph=is_template_graph, update_noise=update_noise, **kwargs)\n",
        "\n",
        "    # Return requested outputs.\n",
        "    images_out = tf.transpose(images_out, [0,2,3,1])\n",
        "    images_out =  tf.reshape(images_out, [-1, DATA_DIM])\n",
        "    if return_dlatents:\n",
        "        return images_out, dlatents\n",
        "    return images_out\n",
        "\n",
        "#################################################################################################\n",
        "\n",
        "def D_stylegan2(\n",
        "    ID,\n",
        "    inputs,                          # First input: Images [minibatch, channel, height, width].\n",
        "    labels,                          # Second input: Labels [minibatch, label_size].\n",
        "    kp1,\n",
        "    kp2,\n",
        "    kp3,                       # Second input: Labels [minibatch, label_size].\n",
        "    NUM_CLASSES=10,\n",
        "    ADA=False,\n",
        "    num_channels        = 3,            # Number of input color channels. Overridden based on dataset.\n",
        "    resolution          = 32,         # Input resolution. Overridden based on dataset.\n",
        "    label_size          = 0,            # Dimensionality of the labels, 0 if no labels. Overridden based on dataset.\n",
        "    fmap_base           = 16 << 10,     # Overall multiplier for the number of feature maps.\n",
        "    fmap_decay          = 1.0,          # log2 feature map reduction when doubling the resolution.\n",
        "    fmap_min            = 1,            # Minimum number of feature maps in any layer.\n",
        "    fmap_max            = 512,          # Maximum number of feature maps in any layer.\n",
        "    architecture        = 'skip',     # Architecture: 'orig', 'skip', 'resnet'.\n",
        "    nonlinearity        = 'lrelu',      # Activation function: 'relu', 'lrelu', etc.\n",
        "    mbstd_group_size    = 32,            # Group size for the minibatch standard deviation layer, 0 = disable.\n",
        "    mbstd_num_features  = 1,            # Number of features for the minibatch standard deviation layer.\n",
        "    mapping_layers      = 0,            # Number of additional mapping layers for the conditioning labels.\n",
        "    mapping_fmaps       = 512,         # Number of activations in the mapping layers, None = default.\n",
        "    mapping_lrmul       = 0.1,          # Learning rate multiplier for the mapping layers.\n",
        "    dtype               = dtype,    # Data type to use for activations and outputs.\n",
        "    resample_kernel     = [1,3,3,1],    # Low-pass filter to apply when resampling activations. None = no filtering.\n",
        "    p=None,\n",
        "    **_kwargs):                         # Ignore unrecognized keyword args.\n",
        "    name='Discriminator%i_'%(ID)\n",
        "    label_size=NUM_CLASSES\n",
        "    labels_in=tf.one_hot(labels, NUM_CLASSES)\n",
        "    images_in = tf.cast(inputs, dtype)\n",
        "    labels_in = tf.cast(labels_in, dtype)\n",
        "    images_in=tf.reshape(images_in, [-1, resolution, resolution, num_channels])\n",
        "    labels_in.set_shape([None, label_size])\n",
        "    def nf(stage): return np.clip(int(fmap_base / (2.0 ** (stage * fmap_decay))), fmap_min, fmap_max)\n",
        "    if mapping_fmaps is None:\n",
        "        mapping_fmaps = nf(0)\n",
        "    # Label embedding and mapping.\n",
        "    if label_size > 0:\n",
        "        y = labels_in\n",
        "        y = apply_bias_act(name+'.LabelEmbed.bias', dense_layer(name+'.LabelEmbed.dense', y, fmaps=mapping_fmaps))\n",
        "        y = normalize_2nd_moment(y)\n",
        "        for idx in range(mapping_layers):\n",
        "            y = apply_bias_act(name+f'.Mapping{idx}.bias',dense_layer(name+f'.Mapping{idx}.dense', y, fmaps=mapping_fmaps, lrmul=mapping_lrmul), act=act, lrmul=mapping_lrmul)\n",
        "        labels_in = y\n",
        "    if ADA:\n",
        "        images_in = augment(images_in,p=p,flip=False,imsp_aug=False)\n",
        "    images_in=tf.transpose(images_in,[0,3,1,2])\n",
        "    resolution=HEIGHT\n",
        "    resolution_log2 = int(np.log2(resolution))\n",
        "    assert resolution == 2**resolution_log2 and resolution >= 4\n",
        "    def nf(stage): return np.clip(int(fmap_base / (2.0 ** (stage * fmap_decay))), fmap_min, fmap_max)\n",
        "    assert architecture in ['orig', 'skip', 'resnet']\n",
        "    act = nonlinearity\n",
        "\n",
        "    # Building blocks for main layers.\n",
        "    def fromrgb(name, x, y, res): # res = 2..resolution_log2\n",
        "        name=name+'_lod%d' % (resolution_log2 - res)\n",
        "        t = apply_bias_act(name+'.Bias', conv2d_layer(name+'.Conv', y, fmaps=nf(res-1), kernel=1), act=act)\n",
        "        return t if x is None else x + t\n",
        "    def block(name, x, res): # res = 2..resolution_log2\n",
        "        t = x\n",
        "        x = apply_bias_act(name+'.Conv0.Bias', conv2d_layer(name+'.Conv0.Conv', x, fmaps=nf(res-1), kernel=3), act=act)\n",
        "        x = apply_bias_act(name+'.Conv1_down.Bias', conv2d_layer(name+'.Conv1_down.Conv', x, fmaps=nf(res-2), kernel=3, down=True, resample_kernel=resample_kernel), act=act)\n",
        "        if architecture == 'resnet':\n",
        "                t = conv2d_layer(name+'.Skip', t, fmaps=nf(res-2), kernel=1, down=True, resample_kernel=resample_kernel)\n",
        "                x = (x + t) * (1 / np.sqrt(2))\n",
        "        return x\n",
        "    def downsample(y):\n",
        "            return downsample_2d(y, k=resample_kernel)\n",
        "\n",
        "    # Main layers.\n",
        "    x = None\n",
        "    y = images_in\n",
        "    for res in range(resolution_log2, 2, -1):\n",
        "            if architecture == 'skip' or res == resolution_log2:\n",
        "                x = fromrgb(name+'.FromRGB%dx%d' % (2**res, 2**res), x, y, res)\n",
        "            x = block(name+'.DB%dx%d'% (2**res, 2**res), x, res)\n",
        "            print('After DBlock%dx%d:' % (2**res, 2**res), x)\n",
        "            if architecture == 'skip':\n",
        "                y = downsample(y)\n",
        "\n",
        "    # Final layers.\n",
        "    if architecture == 'skip':\n",
        "        x = fromrgb(name+'.FromRGB4x4', x, y, 2)\n",
        "    if mbstd_group_size > 1:\n",
        "        x = minibatch_stddev_layer(x, mbstd_group_size, mbstd_num_features)\n",
        "    x = apply_bias_act(name+'.LastConv.Bias', conv2d_layer(name+'.LastConv.Conv', x, fmaps=nf(1), kernel=3), act=act)\n",
        "    feature=x\n",
        "    print('feature:', feature)\n",
        "    x = apply_bias_act(name+'.Dense0.Bias',dense_layer(name+'.Dense0.Conv', x, fmaps=nf(0)), act=act)\n",
        "\n",
        "    x = apply_bias_act(name+'.Output_bias', dense_layer(name+'.Output_dense', x, fmaps=mapping_fmaps if CONDITIONAL else 1))\n",
        "    if CONDITIONAL:\n",
        "        x = tf.reduce_sum(x * labels_in, axis=1, keepdims=True) / np.sqrt(mapping_fmaps)\n",
        "    scores_out = x\n",
        "    print('d output:',x)\n",
        "\n",
        "    # Output.\n",
        "    assert scores_out.dtype == tf.as_dtype(dtype)\n",
        "    return tf.cast(scores_out, tf.float32), tf.cast(feature, tf.float32), None\n",
        "\n",
        "  \n",
        "if STYLEGAN:\n",
        "    Generator = functools.partial(G_main, synthesis_func= 'G_synthesis_stylegan_revised', resolution = HEIGHT, NUM_CLASSES=NUM_CLASSES)\n",
        "    Discriminator=functools.partial(D_stylegan, NUM_CLASSES=NUM_CLASSES, ADA=ADA)\n",
        "elif STYLEGAN2:\n",
        "    Generator = functools.partial(G_main, synthesis_func= 'G_synthesis_stylegan2', resolution = HEIGHT, NUM_CLASSES=NUM_CLASSES)\n",
        "    Discriminator = functools.partial(D_stylegan2, NUM_CLASSES=NUM_CLASSES, ADA=ADA)\n"
      ],
      "execution_count": 18,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "CyNmSZxGJQQ9"
      },
      "source": [
        "train_gen = inf_gen('TRAIN')\n",
        "test_gen = inf_gen('TEST')\n",
        "gen={'training_set':train_gen,'test_set':test_gen}\n",
        "data_set_size={'training_set':TRAIN_SIZE,'test_set':TEST_SIZE}"
      ],
      "execution_count": 19,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "vKet1jL484GR"
      },
      "source": [
        "def chosen_minus_max_unchosen(logits, chosen_idx):\n",
        "    N=int(logits.shape[1])\n",
        "    chosen_logits=tf.boolean_mask(logits, tf.one_hot(chosen_idx, N))\n",
        "    unchosen_logits=tf.reshape(tf.boolean_mask(logits, 1-tf.one_hot(chosen_idx, N)),[-1,N-1]) \n",
        "    max_unchosen_logits=tf.reduce_max(unchosen_logits,1)\n",
        "    return chosen_logits-max_unchosen_logits"
      ],
      "execution_count": 20,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "n5eX39TuJQRB",
        "outputId": "ec0dde52-4a26-44d1-fd65-7c874c2fd11d",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        }
      },
      "source": [
        "# data tensors definitions\n",
        "with tf.device('/cpu:0') and session.as_default():\n",
        "    g_alpha=lib.get_param(name='Generator_alpha', shape=[NUM_GENERATORS], dtype=dtype, initializer=tf.ones_initializer(), trainable=False)\n",
        "    g_w=tf.nn.softmax(g_alpha)\n",
        "    d_alpha=lib.get_param(name='Discriminator_alpha',shape=[NUM_DISCRIMINATORS], dtype=dtype, initializer=tf.ones_initializer(), trainable=False)\n",
        "    d_w=tf.nn.softmax(d_alpha)\n",
        "    d_counter= lib.get_param(name='Discriminator_d_counter',shape=[],dtype=tf.int32, initializer=tf.zeros_initializer(),trainable=False)\n",
        "    indicator= lib.get_param(name='Discriminator_indicator',shape=[4],initializer=tf.zeros_initializer(),trainable=False)\n",
        "    d_p= lib.get_param(name='Discriminator_p',shape=[],initializer=tf.zeros_initializer(),trainable=False)\n",
        "    real_data_splits=tf.split(real_data, num_or_size_splits=NUM_DISCRIMINATORS)\n",
        "    real_labels_splits=tf.split(real_labels, num_or_size_splits=NUM_DISCRIMINATORS)\n",
        "    #real_indices=[tf.where(tf.equal(real_labels%NUM_DISCRIMINATORS, j)) for j in range(NUM_DISCRIMINATORS)]\n",
        "    #real_data_splits=[tf.gather(real_data, real_indices[j]) for j in range(NUM_DISCRIMINATORS)]\n",
        "    #real_labels_splits=[tf.gather(real_labels, real_indices[j]) for j in range(NUM_DISCRIMINATORS)]\n",
        "    \n",
        "    all_fake_sample={}\n",
        "    all_fake_sample_ema={}\n",
        "    #fake_labels=tf.concat([tf.stack([i for i in range(NUM_CLASSES)]) for b in range(BATCH_SIZE*NUM_GENERATORS//NUM_CLASSES)],0)\n",
        "    #fake_labels_splits=tf.split(fake_labels, num_or_size_splits=NUM_GENERATORS)\n",
        "    if RAND_LABELS:\n",
        "      fake_labels=[(tf.random.uniform([BATCH_SIZE],0,max(1,NUM_CLASSES//NUM_GENERATORS),tf.int32)*NUM_GENERATORS+i)%NUM_CLASSES for i in range(NUM_GENERATORS)]\n",
        "    else:\n",
        "      fake_labels=[(tf.range(BATCH_SIZE)*NUM_GENERATORS+i)%NUM_CLASSES for i in range(NUM_GENERATORS)]\n",
        "    stacked_labels=tf.reshape(tf.transpose(tf.stack([fake_labels[i] for i in range(NUM_GENERATORS)]),[1,0]),[-1])\n",
        "    fake_labels_splits=tf.split(stacked_labels, num_or_size_splits=NUM_DISCRIMINATORS)\n",
        "    #fake_labels_splits=real_labels_splits\n",
        "    \n",
        "    \n",
        "\n",
        "    \n",
        "for i in range(NUM_GENERATORS):\n",
        "    with tf.device(DEVICES[i%len(gpu_id)]):\n",
        "        noise = tf.random_normal([BATCH_SIZE, Z_DIM],dtype=dtype)\n",
        "        all_fake_sample[i]=Generator(i,noise,BATCH_SIZE,fake_labels[i], is_training=True)\n",
        "        all_fake_sample_ema[i]=Generator(i,noise,BATCH_SIZE,fake_labels[i],ema=EMA, is_training=False)\n",
        "        #noise = tf.random_normal([int(fake_labels_splits[i].shape[0]), Z_DIM],dtype=dtype)\n",
        "        #all_fake_sample[i]=Generator(i,noise,BATCH_SIZE,fake_labels_splits[i])\n",
        "       \n",
        "\n",
        "with tf.device('/cpu:0'):\n",
        "    stacked_samples=tf.reshape(tf.transpose(tf.stack([all_fake_sample[i] for i in range(NUM_GENERATORS)]),[1,0,2]),[NUM_GENERATORS*BATCH_SIZE,DATA_DIM])\n",
        "    #stacked_samples=tf.concat([tf.stack([all_fake_sample[i][b] for i in range(NUM_GENERATORS)]) for b in range(BATCH_SIZE)],0)\n",
        "    sample_weights_splits=[tf.gather(g_w, fake_labels[i]) for i in range(NUM_GENERATORS)]\n",
        "    fake_sample=tf.gather(stacked_samples,tf.range(min(100,stacked_samples.shape[0])))\n",
        "    fake_sample_splits=tf.split(stacked_samples, num_or_size_splits=NUM_DISCRIMINATORS)\n",
        "    #fake_sample_splits=all_fake_sample\n",
        "    \n",
        "    stacked_samples_ema=tf.reshape(tf.transpose(tf.stack([all_fake_sample_ema[i] for i in range(NUM_GENERATORS)]),[1,0,2]),[NUM_GENERATORS*BATCH_SIZE,DATA_DIM])\n",
        "    fake_sample=tf.gather(stacked_samples_ema,tf.range(min(100,stacked_samples.shape[0])))\n",
        "    fake_sample=tf.clip_by_value(fake_sample, -1., 1.)\n",
        "disc_real=[]\n",
        "log_sigmoid_disc_real=[]\n",
        "indicator_stat=[]\n",
        "disc_real_hinge=[]\n",
        "ac_real={}\n",
        "disc_fake=[]\n",
        "disc_fake_hinge=[]\n",
        "log_1_minus_sigmoid_disc_fake=[]\n",
        "log_sigmoid_disc_fake=[]\n",
        "ac_fake={}\n",
        "disc_real_noisy_1,disc_real_features_noisy_1, ac_real_noisy_1={},{},{}\n",
        "disc_real_noisy_2,disc_real_features_noisy_2, ac_real_noisy_2={},{},{}\n",
        "disc_fake_noisy,disc_fake_features_noisy, ac_fake_noisy={},{},{}\n",
        "\n",
        "for j in range(NUM_DISCRIMINATORS):\n",
        "    with tf.device(DEVICES[j%len(gpu_id)]):\n",
        "        # x' with dropout noise added\n",
        "        disc_real_noisy_1[j],disc_real_features_noisy_1[j], ac_real_noisy_1[j] = Discriminator(j, real_data_splits[j], real_labels_splits[j],0.8,0.5,0.5)  #dropout rate of 0.2,0.5,0.5\n",
        "        # x'' with dropout noise added\n",
        "        if LAMBDA2: \n",
        "          disc_real_noisy_2[j],disc_real_features_noisy_2[j], ac_real_noisy_2[j] = Discriminator(j, real_data_splits[j], real_labels_splits[j],0.8,0.5,0.5)  #dropout rate of 0.2,0.5,0.5\n",
        "\n",
        "        disc_real.append(d_w[j]*tf.reduce_mean(disc_real_noisy_1[j]))\n",
        "        log_sigmoid_disc_real.append(d_w[j]*tf.reduce_mean(tf.nn.softplus(-disc_real_noisy_1[j])))\n",
        "        disc_real_hinge.append(d_w[j]*tf.reduce_mean(-tf.nn.relu(1-disc_real_noisy_1[j])))\n",
        "        # fake data with dropout noise added\n",
        "        disc_fake_noisy[j],disc_fake_features_noisy[j], ac_fake_noisy[j] = Discriminator(j,fake_sample_splits[j],fake_labels_splits[j], 0.8,0.5,0.5)  #dropout rate of 0.2,0.5,0.5\n",
        "        #disc_fake+=d_w[j]*sample_weights_splits*disc_fake_noisy[j]\n",
        "        disc_fake.append(d_w[j]*tf.reduce_mean(disc_fake_noisy[j]))\n",
        "        disc_fake_hinge.append(d_w[j]*tf.reduce_mean(tf.nn.relu(1+disc_fake_noisy[j])))\n",
        "        log_1_minus_sigmoid_disc_fake.append(d_w[j]*tf.reduce_mean(tf.nn.softplus(disc_fake_noisy[j])))\n",
        "        log_sigmoid_disc_fake.append(d_w[j]*tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(labels=tf.ones_like(disc_fake_noisy[j]),logits=disc_fake_noisy[j])))\n",
        "        if CONDITIONAL and ACGAN:\n",
        "            ac_real[j]=ac_real_noisy_1[j]\n",
        "            ac_fake[j]=ac_fake_noisy[j]\n",
        "        if MH_LOSS:\n",
        "            indicator_stat.append(d_w[j]*tf.reduce_mean(chosen_minus_max_unchosen(ac_fake_noisy[j],NUM_CLASSES*tf.ones_like(real_labels_splits[j]))))\n",
        "        elif AM_LOSS:\n",
        "            indicator_stat.append(d_w[j]*tf.reduce_mean(tf.math.sign(tf.nn.softmax(ac_fake_noisy[j])[:,NUM_CLASSES]-.5)))\n",
        "        else:\n",
        "            indicator_stat.append(d_w[j]*tf.reduce_mean(tf.math.sign(disc_real_noisy_1[j])))\n",
        "# not applicable to WGAN:\n",
        "with tf.device('/cpu:0'):\n",
        "    indicator_stat=tf.add_n(indicator_stat)\n",
        "    disc_real_accuracy = tf.reduce_mean([tf.cast(disc_real[j]>=0, dtype) for j in range(NUM_DISCRIMINATORS)])\n",
        "    disc_fake_accuracy = tf.reduce_mean([tf.cast(disc_fake[j]<=0, dtype) for j in range(NUM_DISCRIMINATORS)])\n",
        "    disc_accuracy=(disc_real_accuracy+disc_fake_accuracy)/2\n",
        "    disc_real=tf.add_n(disc_real)\n",
        "    disc_real_hinge=tf.add_n(disc_real_hinge)\n",
        "    disc_fake=tf.add_n(disc_fake)\n",
        "    disc_fake_hinge=tf.add_n(disc_fake_hinge)\n",
        "    log_sigmoid_disc_real=tf.add_n(log_sigmoid_disc_real)\n",
        "    log_1_minus_sigmoid_disc_fake=tf.add_n(log_1_minus_sigmoid_disc_fake)\n",
        "    log_sigmoid_disc_fake=tf.add_n(log_sigmoid_disc_fake)"
      ],
      "execution_count": 21,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Tensor(\"one_hot_1:0\", shape=(50, 10), dtype=float32, device=/device:GPU:0)\n",
            "code:  [<tf.Tensor 'concat_2:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_3:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_4:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_5:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_6:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_7:0' shape=(50, 148) dtype=float32>]\n",
            "after dense Tensor(\"Reshape_3:0\", shape=(50, 4, 4, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 0: Tensor(\"add_14:0\", shape=(50, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 1: Tensor(\"add_17:0\", shape=(50, 16, 16, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 2: Tensor(\"add_20:0\", shape=(50, 32, 32, 256), dtype=float32, device=/device:GPU:0)\n",
            "Tensor(\"one_hot_3:0\", shape=(50, 10), dtype=float32, device=/device:GPU:0)\n",
            "code:  [<tf.Tensor 'concat_8:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_9:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_10:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_11:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_12:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_13:0' shape=(50, 148) dtype=float32>]\n",
            "after dense Tensor(\"Reshape_63:0\", shape=(50, 4, 4, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 0: Tensor(\"add_24:0\", shape=(50, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 1: Tensor(\"add_27:0\", shape=(50, 16, 16, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 2: Tensor(\"add_30:0\", shape=(50, 32, 32, 256), dtype=float32, device=/device:GPU:0)\n",
            "Tensor(\"one_hot_5:0\", shape=(50, 10), dtype=float32, device=/device:GPU:0)\n",
            "code:  [<tf.Tensor 'concat_14:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_15:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_16:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_17:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_18:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_19:0' shape=(50, 148) dtype=float32>]\n",
            "after dense Tensor(\"Reshape_123:0\", shape=(50, 4, 4, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 0: Tensor(\"add_34:0\", shape=(50, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 1: Tensor(\"add_37:0\", shape=(50, 16, 16, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 2: Tensor(\"add_40:0\", shape=(50, 32, 32, 256), dtype=float32, device=/device:GPU:0)\n",
            "Tensor(\"one_hot_7:0\", shape=(50, 10), dtype=float32, device=/device:GPU:0)\n",
            "code:  [<tf.Tensor 'concat_20:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_21:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_22:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_23:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_24:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_25:0' shape=(50, 148) dtype=float32>]\n",
            "after dense Tensor(\"Reshape_183:0\", shape=(50, 4, 4, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 0: Tensor(\"add_44:0\", shape=(50, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 1: Tensor(\"add_47:0\", shape=(50, 16, 16, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 2: Tensor(\"add_50:0\", shape=(50, 32, 32, 256), dtype=float32, device=/device:GPU:0)\n",
            "Tensor(\"one_hot_9:0\", shape=(50, 10), dtype=float32, device=/device:GPU:0)\n",
            "code:  [<tf.Tensor 'concat_26:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_27:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_28:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_29:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_30:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_31:0' shape=(50, 148) dtype=float32>]\n",
            "after dense Tensor(\"Reshape_243:0\", shape=(50, 4, 4, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 0: Tensor(\"add_54:0\", shape=(50, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 1: Tensor(\"add_57:0\", shape=(50, 16, 16, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 2: Tensor(\"add_60:0\", shape=(50, 32, 32, 256), dtype=float32, device=/device:GPU:0)\n",
            "Tensor(\"one_hot_11:0\", shape=(50, 10), dtype=float32, device=/device:GPU:0)\n",
            "code:  [<tf.Tensor 'concat_32:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_33:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_34:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_35:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_36:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_37:0' shape=(50, 148) dtype=float32>]\n",
            "after dense Tensor(\"Reshape_303:0\", shape=(50, 4, 4, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 0: Tensor(\"add_64:0\", shape=(50, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 1: Tensor(\"add_67:0\", shape=(50, 16, 16, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 2: Tensor(\"add_70:0\", shape=(50, 32, 32, 256), dtype=float32, device=/device:GPU:0)\n",
            "Tensor(\"one_hot_13:0\", shape=(50, 10), dtype=float32, device=/device:GPU:0)\n",
            "code:  [<tf.Tensor 'concat_38:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_39:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_40:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_41:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_42:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_43:0' shape=(50, 148) dtype=float32>]\n",
            "after dense Tensor(\"Reshape_363:0\", shape=(50, 4, 4, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 0: Tensor(\"add_74:0\", shape=(50, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 1: Tensor(\"add_77:0\", shape=(50, 16, 16, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 2: Tensor(\"add_80:0\", shape=(50, 32, 32, 256), dtype=float32, device=/device:GPU:0)\n",
            "Tensor(\"one_hot_15:0\", shape=(50, 10), dtype=float32, device=/device:GPU:0)\n",
            "code:  [<tf.Tensor 'concat_44:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_45:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_46:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_47:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_48:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_49:0' shape=(50, 148) dtype=float32>]\n",
            "after dense Tensor(\"Reshape_423:0\", shape=(50, 4, 4, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 0: Tensor(\"add_84:0\", shape=(50, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 1: Tensor(\"add_87:0\", shape=(50, 16, 16, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 2: Tensor(\"add_90:0\", shape=(50, 32, 32, 256), dtype=float32, device=/device:GPU:0)\n",
            "Tensor(\"one_hot_17:0\", shape=(50, 10), dtype=float32, device=/device:GPU:0)\n",
            "code:  [<tf.Tensor 'concat_50:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_51:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_52:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_53:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_54:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_55:0' shape=(50, 148) dtype=float32>]\n",
            "after dense Tensor(\"Reshape_483:0\", shape=(50, 4, 4, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 0: Tensor(\"add_94:0\", shape=(50, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 1: Tensor(\"add_97:0\", shape=(50, 16, 16, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 2: Tensor(\"add_100:0\", shape=(50, 32, 32, 256), dtype=float32, device=/device:GPU:0)\n",
            "Tensor(\"one_hot_19:0\", shape=(50, 10), dtype=float32, device=/device:GPU:0)\n",
            "code:  [<tf.Tensor 'concat_56:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_57:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_58:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_59:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_60:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_61:0' shape=(50, 148) dtype=float32>]\n",
            "after dense Tensor(\"Reshape_543:0\", shape=(50, 4, 4, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 0: Tensor(\"add_104:0\", shape=(50, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 1: Tensor(\"add_107:0\", shape=(50, 16, 16, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 2: Tensor(\"add_110:0\", shape=(50, 32, 32, 256), dtype=float32, device=/device:GPU:0)\n",
            "Tensor(\"one_hot_21:0\", shape=(50, 10), dtype=float32, device=/device:GPU:0)\n",
            "code:  [<tf.Tensor 'concat_62:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_63:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_64:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_65:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_66:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_67:0' shape=(50, 148) dtype=float32>]\n",
            "after dense Tensor(\"Reshape_603:0\", shape=(50, 4, 4, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 0: Tensor(\"add_114:0\", shape=(50, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 1: Tensor(\"add_117:0\", shape=(50, 16, 16, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 2: Tensor(\"add_120:0\", shape=(50, 32, 32, 256), dtype=float32, device=/device:GPU:0)\n",
            "Tensor(\"one_hot_23:0\", shape=(50, 10), dtype=float32, device=/device:GPU:0)\n",
            "code:  [<tf.Tensor 'concat_68:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_69:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_70:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_71:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_72:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_73:0' shape=(50, 148) dtype=float32>]\n",
            "after dense Tensor(\"Reshape_663:0\", shape=(50, 4, 4, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 0: Tensor(\"add_124:0\", shape=(50, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 1: Tensor(\"add_127:0\", shape=(50, 16, 16, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 2: Tensor(\"add_130:0\", shape=(50, 32, 32, 256), dtype=float32, device=/device:GPU:0)\n",
            "Tensor(\"one_hot_25:0\", shape=(50, 10), dtype=float32, device=/device:GPU:0)\n",
            "code:  [<tf.Tensor 'concat_74:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_75:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_76:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_77:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_78:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_79:0' shape=(50, 148) dtype=float32>]\n",
            "after dense Tensor(\"Reshape_723:0\", shape=(50, 4, 4, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 0: Tensor(\"add_134:0\", shape=(50, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 1: Tensor(\"add_137:0\", shape=(50, 16, 16, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 2: Tensor(\"add_140:0\", shape=(50, 32, 32, 256), dtype=float32, device=/device:GPU:0)\n",
            "Tensor(\"one_hot_27:0\", shape=(50, 10), dtype=float32, device=/device:GPU:0)\n",
            "code:  [<tf.Tensor 'concat_80:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_81:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_82:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_83:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_84:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_85:0' shape=(50, 148) dtype=float32>]\n",
            "after dense Tensor(\"Reshape_783:0\", shape=(50, 4, 4, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 0: Tensor(\"add_144:0\", shape=(50, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 1: Tensor(\"add_147:0\", shape=(50, 16, 16, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 2: Tensor(\"add_150:0\", shape=(50, 32, 32, 256), dtype=float32, device=/device:GPU:0)\n",
            "Tensor(\"one_hot_29:0\", shape=(50, 10), dtype=float32, device=/device:GPU:0)\n",
            "code:  [<tf.Tensor 'concat_86:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_87:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_88:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_89:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_90:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_91:0' shape=(50, 148) dtype=float32>]\n",
            "after dense Tensor(\"Reshape_843:0\", shape=(50, 4, 4, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 0: Tensor(\"add_154:0\", shape=(50, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 1: Tensor(\"add_157:0\", shape=(50, 16, 16, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 2: Tensor(\"add_160:0\", shape=(50, 32, 32, 256), dtype=float32, device=/device:GPU:0)\n",
            "Tensor(\"one_hot_31:0\", shape=(50, 10), dtype=float32, device=/device:GPU:0)\n",
            "code:  [<tf.Tensor 'concat_92:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_93:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_94:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_95:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_96:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_97:0' shape=(50, 148) dtype=float32>]\n",
            "after dense Tensor(\"Reshape_903:0\", shape=(50, 4, 4, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 0: Tensor(\"add_164:0\", shape=(50, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 1: Tensor(\"add_167:0\", shape=(50, 16, 16, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 2: Tensor(\"add_170:0\", shape=(50, 32, 32, 256), dtype=float32, device=/device:GPU:0)\n",
            "Tensor(\"one_hot_33:0\", shape=(50, 10), dtype=float32, device=/device:GPU:0)\n",
            "code:  [<tf.Tensor 'concat_98:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_99:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_100:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_101:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_102:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_103:0' shape=(50, 148) dtype=float32>]\n",
            "after dense Tensor(\"Reshape_963:0\", shape=(50, 4, 4, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 0: Tensor(\"add_174:0\", shape=(50, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 1: Tensor(\"add_177:0\", shape=(50, 16, 16, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 2: Tensor(\"add_180:0\", shape=(50, 32, 32, 256), dtype=float32, device=/device:GPU:0)\n",
            "Tensor(\"one_hot_35:0\", shape=(50, 10), dtype=float32, device=/device:GPU:0)\n",
            "code:  [<tf.Tensor 'concat_104:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_105:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_106:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_107:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_108:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_109:0' shape=(50, 148) dtype=float32>]\n",
            "after dense Tensor(\"Reshape_1023:0\", shape=(50, 4, 4, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 0: Tensor(\"add_184:0\", shape=(50, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 1: Tensor(\"add_187:0\", shape=(50, 16, 16, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 2: Tensor(\"add_190:0\", shape=(50, 32, 32, 256), dtype=float32, device=/device:GPU:0)\n",
            "Tensor(\"one_hot_37:0\", shape=(50, 10), dtype=float32, device=/device:GPU:0)\n",
            "code:  [<tf.Tensor 'concat_110:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_111:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_112:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_113:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_114:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_115:0' shape=(50, 148) dtype=float32>]\n",
            "after dense Tensor(\"Reshape_1083:0\", shape=(50, 4, 4, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 0: Tensor(\"add_194:0\", shape=(50, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 1: Tensor(\"add_197:0\", shape=(50, 16, 16, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 2: Tensor(\"add_200:0\", shape=(50, 32, 32, 256), dtype=float32, device=/device:GPU:0)\n",
            "Tensor(\"one_hot_39:0\", shape=(50, 10), dtype=float32, device=/device:GPU:0)\n",
            "code:  [<tf.Tensor 'concat_116:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_117:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_118:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_119:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_120:0' shape=(50, 148) dtype=float32>, <tf.Tensor 'concat_121:0' shape=(50, 148) dtype=float32>]\n",
            "after dense Tensor(\"Reshape_1143:0\", shape=(50, 4, 4, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 0: Tensor(\"add_204:0\", shape=(50, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 1: Tensor(\"add_207:0\", shape=(50, 16, 16, 256), dtype=float32, device=/device:GPU:0)\n",
            "After Block 2: Tensor(\"add_210:0\", shape=(50, 32, 32, 256), dtype=float32, device=/device:GPU:0)\n",
            "inputs: Tensor(\"Reshape_1203:0\", shape=(?, 32, 32, 3), dtype=float32, device=/device:GPU:0)\n",
            "after resdown1 Tensor(\"add_212:0\", shape=(?, 16, 16, 256), dtype=float32, device=/device:GPU:0)\n",
            "after resdown2 Tensor(\"add_213:0\", shape=(?, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "after res8 Tensor(\"add_214:0\", shape=(?, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "after res16 Tensor(\"add_215:0\", shape=(?, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "inputs: Tensor(\"Reshape_1231:0\", shape=(50, 32, 32, 3), dtype=float32, device=/device:GPU:0)\n",
            "after resdown1 Tensor(\"add_217:0\", shape=(50, 16, 16, 256), dtype=float32, device=/device:GPU:0)\n",
            "after resdown2 Tensor(\"add_218:0\", shape=(50, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "after res8 Tensor(\"add_219:0\", shape=(50, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "after res16 Tensor(\"add_220:0\", shape=(50, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "inputs: Tensor(\"Reshape_1260:0\", shape=(?, 32, 32, 3), dtype=float32, device=/device:GPU:0)\n",
            "after resdown1 Tensor(\"add_223:0\", shape=(?, 16, 16, 256), dtype=float32, device=/device:GPU:0)\n",
            "after resdown2 Tensor(\"add_224:0\", shape=(?, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "after res8 Tensor(\"add_225:0\", shape=(?, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "after res16 Tensor(\"add_226:0\", shape=(?, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "inputs: Tensor(\"Reshape_1288:0\", shape=(50, 32, 32, 3), dtype=float32, device=/device:GPU:0)\n",
            "after resdown1 Tensor(\"add_228:0\", shape=(50, 16, 16, 256), dtype=float32, device=/device:GPU:0)\n",
            "after resdown2 Tensor(\"add_229:0\", shape=(50, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "after res8 Tensor(\"add_230:0\", shape=(50, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "after res16 Tensor(\"add_231:0\", shape=(50, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "inputs: Tensor(\"Reshape_1317:0\", shape=(?, 32, 32, 3), dtype=float32, device=/device:GPU:0)\n",
            "after resdown1 Tensor(\"add_234:0\", shape=(?, 16, 16, 256), dtype=float32, device=/device:GPU:0)\n",
            "after resdown2 Tensor(\"add_235:0\", shape=(?, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "after res8 Tensor(\"add_236:0\", shape=(?, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "after res16 Tensor(\"add_237:0\", shape=(?, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "inputs: Tensor(\"Reshape_1345:0\", shape=(50, 32, 32, 3), dtype=float32, device=/device:GPU:0)\n",
            "after resdown1 Tensor(\"add_239:0\", shape=(50, 16, 16, 256), dtype=float32, device=/device:GPU:0)\n",
            "after resdown2 Tensor(\"add_240:0\", shape=(50, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "after res8 Tensor(\"add_241:0\", shape=(50, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "after res16 Tensor(\"add_242:0\", shape=(50, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "inputs: Tensor(\"Reshape_1374:0\", shape=(?, 32, 32, 3), dtype=float32, device=/device:GPU:0)\n",
            "after resdown1 Tensor(\"add_245:0\", shape=(?, 16, 16, 256), dtype=float32, device=/device:GPU:0)\n",
            "after resdown2 Tensor(\"add_246:0\", shape=(?, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "after res8 Tensor(\"add_247:0\", shape=(?, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "after res16 Tensor(\"add_248:0\", shape=(?, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "inputs: Tensor(\"Reshape_1402:0\", shape=(50, 32, 32, 3), dtype=float32, device=/device:GPU:0)\n",
            "after resdown1 Tensor(\"add_250:0\", shape=(50, 16, 16, 256), dtype=float32, device=/device:GPU:0)\n",
            "after resdown2 Tensor(\"add_251:0\", shape=(50, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "after res8 Tensor(\"add_252:0\", shape=(50, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "after res16 Tensor(\"add_253:0\", shape=(50, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "inputs: Tensor(\"Reshape_1431:0\", shape=(?, 32, 32, 3), dtype=float32, device=/device:GPU:0)\n",
            "after resdown1 Tensor(\"add_256:0\", shape=(?, 16, 16, 256), dtype=float32, device=/device:GPU:0)\n",
            "after resdown2 Tensor(\"add_257:0\", shape=(?, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "after res8 Tensor(\"add_258:0\", shape=(?, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "after res16 Tensor(\"add_259:0\", shape=(?, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "inputs: Tensor(\"Reshape_1459:0\", shape=(50, 32, 32, 3), dtype=float32, device=/device:GPU:0)\n",
            "after resdown1 Tensor(\"add_261:0\", shape=(50, 16, 16, 256), dtype=float32, device=/device:GPU:0)\n",
            "after resdown2 Tensor(\"add_262:0\", shape=(50, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "after res8 Tensor(\"add_263:0\", shape=(50, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "after res16 Tensor(\"add_264:0\", shape=(50, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "inputs: Tensor(\"Reshape_1488:0\", shape=(?, 32, 32, 3), dtype=float32, device=/device:GPU:0)\n",
            "after resdown1 Tensor(\"add_267:0\", shape=(?, 16, 16, 256), dtype=float32, device=/device:GPU:0)\n",
            "after resdown2 Tensor(\"add_268:0\", shape=(?, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "after res8 Tensor(\"add_269:0\", shape=(?, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "after res16 Tensor(\"add_270:0\", shape=(?, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "inputs: Tensor(\"Reshape_1516:0\", shape=(50, 32, 32, 3), dtype=float32, device=/device:GPU:0)\n",
            "after resdown1 Tensor(\"add_272:0\", shape=(50, 16, 16, 256), dtype=float32, device=/device:GPU:0)\n",
            "after resdown2 Tensor(\"add_273:0\", shape=(50, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "after res8 Tensor(\"add_274:0\", shape=(50, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "after res16 Tensor(\"add_275:0\", shape=(50, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "inputs: Tensor(\"Reshape_1545:0\", shape=(?, 32, 32, 3), dtype=float32, device=/device:GPU:0)\n",
            "after resdown1 Tensor(\"add_278:0\", shape=(?, 16, 16, 256), dtype=float32, device=/device:GPU:0)\n",
            "after resdown2 Tensor(\"add_279:0\", shape=(?, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "after res8 Tensor(\"add_280:0\", shape=(?, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "after res16 Tensor(\"add_281:0\", shape=(?, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "inputs: Tensor(\"Reshape_1573:0\", shape=(50, 32, 32, 3), dtype=float32, device=/device:GPU:0)\n",
            "after resdown1 Tensor(\"add_283:0\", shape=(50, 16, 16, 256), dtype=float32, device=/device:GPU:0)\n",
            "after resdown2 Tensor(\"add_284:0\", shape=(50, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "after res8 Tensor(\"add_285:0\", shape=(50, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "after res16 Tensor(\"add_286:0\", shape=(50, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "inputs: Tensor(\"Reshape_1602:0\", shape=(?, 32, 32, 3), dtype=float32, device=/device:GPU:0)\n",
            "after resdown1 Tensor(\"add_289:0\", shape=(?, 16, 16, 256), dtype=float32, device=/device:GPU:0)\n",
            "after resdown2 Tensor(\"add_290:0\", shape=(?, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "after res8 Tensor(\"add_291:0\", shape=(?, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "after res16 Tensor(\"add_292:0\", shape=(?, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "inputs: Tensor(\"Reshape_1630:0\", shape=(50, 32, 32, 3), dtype=float32, device=/device:GPU:0)\n",
            "after resdown1 Tensor(\"add_294:0\", shape=(50, 16, 16, 256), dtype=float32, device=/device:GPU:0)\n",
            "after resdown2 Tensor(\"add_295:0\", shape=(50, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "after res8 Tensor(\"add_296:0\", shape=(50, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "after res16 Tensor(\"add_297:0\", shape=(50, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "inputs: Tensor(\"Reshape_1659:0\", shape=(?, 32, 32, 3), dtype=float32, device=/device:GPU:0)\n",
            "after resdown1 Tensor(\"add_300:0\", shape=(?, 16, 16, 256), dtype=float32, device=/device:GPU:0)\n",
            "after resdown2 Tensor(\"add_301:0\", shape=(?, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "after res8 Tensor(\"add_302:0\", shape=(?, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "after res16 Tensor(\"add_303:0\", shape=(?, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "inputs: Tensor(\"Reshape_1687:0\", shape=(50, 32, 32, 3), dtype=float32, device=/device:GPU:0)\n",
            "after resdown1 Tensor(\"add_305:0\", shape=(50, 16, 16, 256), dtype=float32, device=/device:GPU:0)\n",
            "after resdown2 Tensor(\"add_306:0\", shape=(50, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "after res8 Tensor(\"add_307:0\", shape=(50, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "after res16 Tensor(\"add_308:0\", shape=(50, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "inputs: Tensor(\"Reshape_1716:0\", shape=(?, 32, 32, 3), dtype=float32, device=/device:GPU:0)\n",
            "after resdown1 Tensor(\"add_311:0\", shape=(?, 16, 16, 256), dtype=float32, device=/device:GPU:0)\n",
            "after resdown2 Tensor(\"add_312:0\", shape=(?, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "after res8 Tensor(\"add_313:0\", shape=(?, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "after res16 Tensor(\"add_314:0\", shape=(?, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "inputs: Tensor(\"Reshape_1744:0\", shape=(50, 32, 32, 3), dtype=float32, device=/device:GPU:0)\n",
            "after resdown1 Tensor(\"add_316:0\", shape=(50, 16, 16, 256), dtype=float32, device=/device:GPU:0)\n",
            "after resdown2 Tensor(\"add_317:0\", shape=(50, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "after res8 Tensor(\"add_318:0\", shape=(50, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n",
            "after res16 Tensor(\"add_319:0\", shape=(50, 8, 8, 256), dtype=float32, device=/device:GPU:0)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "hAEkOZOX0gRC"
      },
      "source": [
        ""
      ],
      "execution_count": 21,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "dfGoubfnJQRI",
        "outputId": "53062d1a-0a36-4a49-e792-8e8ef17abebf",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 250
        }
      },
      "source": [
        "#cost definitions\n",
        "\n",
        "###################################################### Discriminator #####################################################################\n",
        "with tf.device('/cpu:0'):\n",
        "    disc_cost=tf.zeros([],dtype=dtype)\n",
        "    gen_cost=tf.zeros([],dtype=dtype)\n",
        "    g_orth_reg=tf.zeros([],dtype=dtype)\n",
        "    d_orth_reg=tf.zeros([],dtype=dtype)\n",
        "    # entropy regularizers:\n",
        "    gen_cost -= tf.reduce_mean(tf.log(g_w))\n",
        "    disc_cost -= tf.reduce_mean(tf.log(d_w))\n",
        "\n",
        "    # G LOSS\n",
        "    if GAN_LOSS:\n",
        "        gen_cost+=log_sigmoid_disc_fake\n",
        "    elif HINGE_LOSS or WGAN_LOSS:\n",
        "        gen_cost-=disc_fake\n",
        "    #D LOSS\n",
        "    if GAN_LOSS:\n",
        "        disc_cost+=(log_sigmoid_disc_real+log_1_minus_sigmoid_disc_fake)\n",
        "    elif HINGE_LOSS:\n",
        "        disc_cost+=disc_fake_hinge-disc_real_hinge\n",
        "    elif WGAN_LOSS:\n",
        "        disc_cost+=disc_fake-disc_real\n",
        "# WGAN gradient penalty\n",
        "gradient_penalty=tf.zeros([])\n",
        "wass_dist=tf.zeros([])\n",
        "if COMPUTE_GRADIENT:\n",
        "    gradient_penalty=[]\n",
        "    wass_dist=[]\n",
        "    for j in range(NUM_DISCRIMINATORS):\n",
        "        with tf.device(DEVICES[j%len(gpu_id)]):\n",
        "            eps = tf.random_uniform(\n",
        "            shape=[int(fake_sample_splits[j].shape[0]),1], \n",
        "            minval=0.,\n",
        "            maxval=1.,\n",
        "            dtype=dtype\n",
        "            )\n",
        "            #interpolates = fake_sample_splits[j]\n",
        "            if R1_COEFF:\n",
        "                disc_output=tf.reduce_sum(disc_real_noisy_1[j])\n",
        "                #if len(disc_output.shape)>1 and disc_output.shape[-1]==NUM_CLASSES+1:\n",
        "                #    disc_output=-tf.log(1/tf.reduce_sum(tf.nn.softmax(disc_output)[:,:NUM_CLASSES],1)-1) #logit function of real probability\n",
        "                gradients = tf.gradients(disc_output, [real_data_splits[j]])[0]  #same dropout rate\n",
        "            else:\n",
        "                gradients = tf.gradients(disc_fake_noisy[j], [fake_sample_splits[j]])[0]  #same dropout rate\n",
        "            square_slopes=tf.reduce_sum(tf.square(gradients), reduction_indices=[1])\n",
        "            print('square_slopes:',square_slopes)\n",
        "            slopes = tf.sqrt(square_slopes)   \n",
        "            max_slope=tf.reduce_max(slopes)  \n",
        "            wass_dist.append(d_w[j]*(tf.reduce_mean(disc_real_noisy_1[j])-tf.reduce_mean(disc_fake_noisy[j]))/max_slope)\n",
        "            if LAMBDA:\n",
        "                gradient_penalty.append(d_w[j]*LAMBDA*tf.reduce_mean((slopes-1.)**2))\n",
        "            elif R1_COEFF:\n",
        "                gradient_penalty.append(d_w[j]*R1_COEFF*tf.reduce_mean(square_slopes)/2)\n",
        "    with tf.device('/cpu:0'):\n",
        "        gradient_penalty=tf.add_n(gradient_penalty)\n",
        "        if LAMBDA or R1_COEFF:\n",
        "            disc_cost+=gradient_penalty\n",
        "        wass_dist=tf.reduce_max(wass_dist)\n",
        "\n",
        "# ACGAN Classifier cost for G\n",
        "gen_acgan_cost=[]\n",
        "for j in range(NUM_DISCRIMINATORS):\n",
        "    with tf.device(DEVICES[j%len(gpu_id)]):\n",
        "        if CONDITIONAL and ACGAN: \n",
        "            gen_acgan_cost.append(d_w[j]*ACGAN_SCALE_G*tf.reduce_mean(tf.nn.sparse_softmax_cross_entropy_with_logits(logits=ac_fake[j], labels=fake_labels_splits[j])))\n",
        "with tf.device('/cpu:0'):\n",
        "    if CONDITIONAL and ACGAN: \n",
        "        gen_cost+=tf.add_n(gen_acgan_cost)\n",
        "# ACGAN classification cost\n",
        "ac_cost=[]\n",
        "ac_real_acc=[]\n",
        "ac_fake_acc=[]\n",
        "CT_cost=[]\n",
        "D_MHINGE_LOSS=[]\n",
        "G_MHINGE_LOSS=[]\n",
        "FM_loss=[]\n",
        "for j in range(NUM_DISCRIMINATORS):\n",
        "    with tf.device(DEVICES[j%len(gpu_id)]):\n",
        "        if CONDITIONAL and ACGAN:\n",
        "            # cross-entropy classification cost\n",
        "            ac_cost.append(d_w[j]*ACGAN_SCALE*(tf.reduce_mean(tf.nn.sparse_softmax_cross_entropy_with_logits(logits=ac_real[j], labels=real_labels_splits[j]))))\n",
        "            #auxiliary classifier(in discriminator) accuracy on real data\n",
        "            ac_real_acc.append(d_w[j]*(tf.reduce_mean(tf.cast(tf.equal(tf.to_int32(tf.argmax(ac_real[j], dimension=1)), real_labels_splits[j]),dtype))))\n",
        "            #auxiliary classifier(in discriminator) accuracy on fake data\n",
        "            ac_fake_acc.append(d_w[j]*(tf.reduce_mean(tf.cast(tf.equal(tf.to_int32(tf.argmax(ac_fake[j], dimension=1)),fake_labels_splits[j]),dtype))))\n",
        "        if LAMBDA2:    \n",
        "            # CT-GAN consistency term\n",
        "            CT = LAMBDA2*tf.square(disc_real_noisy_1[j]-disc_real_noisy_2[j])  \n",
        "            CT += LAMBDA2*0.1*tf.reduce_mean(tf.square(disc_real_features_noisy_1[j]-disc_real_features_noisy_2[j]),reduction_indices=[1])\n",
        "            CT_ = tf.maximum(CT-Factor_M,0.)\n",
        "            CT_ = tf.reduce_mean(CT_)\n",
        "            CT_cost.append(d_w[j]*(CT_))\n",
        "        if MH_LOSS or AM_LOSS:\n",
        "            if AM_LOSS:\n",
        "                ac_real_cost=tf.reduce_mean(tf.nn.sparse_softmax_cross_entropy_with_logits(logits=ac_real_noisy_1[j],labels=real_labels_splits[j]))\n",
        "                ac_fake_cost=tf.reduce_mean(tf.nn.sparse_softmax_cross_entropy_with_logits(logits=ac_fake_noisy[j],labels=NUM_CLASSES*tf.ones_like(real_labels_splits[j])))\n",
        "                G_MHINGE_LOSS.append(d_w[j]*tf.reduce_mean(tf.nn.sparse_softmax_cross_entropy_with_logits(logits=ac_fake_noisy[j],labels=fake_labels_splits[j])))\n",
        "            else:\n",
        "                ac_real_cost=tf.reduce_mean(tf.nn.relu(1-chosen_minus_max_unchosen(ac_real_noisy_1[j],real_labels_splits[j])))\n",
        "                ac_fake_cost=tf.reduce_mean(tf.nn.relu(1-chosen_minus_max_unchosen(ac_fake_noisy[j],NUM_CLASSES*tf.ones_like(real_labels_splits[j]))))\n",
        "                G_MHINGE_LOSS.append(d_w[j]*tf.reduce_mean(tf.nn.relu(1-chosen_minus_max_unchosen(ac_fake_noisy[j],fake_labels_splits[j]))))\n",
        "            ac_cost.append(d_w[j]*ac_real_cost)\n",
        "            D_MHINGE_LOSS.append(d_w[j]*(ac_real_cost+ac_fake_cost))\n",
        "            ac_real_acc.append(d_w[j]*(tf.reduce_mean(tf.cast(tf.equal(tf.to_int32(tf.argmax(ac_real_noisy_1[j], dimension=1)), real_labels_splits[j]),dtype))))\n",
        "            ac_fake_acc.append(d_w[j]*(tf.reduce_mean(tf.cast(tf.equal(tf.to_int32(tf.argmax(ac_fake_noisy[j], dimension=1)),fake_labels_splits[j]),dtype))))\n",
        "        if MH_LOSS or AM_LOSS:\n",
        "            FM_loss.append(d_w[j]*tf.reduce_sum(tf.abs(tf.reduce_mean(disc_real_features_noisy_1[j],0)-tf.reduce_mean(disc_fake_features_noisy[j],0))))\n",
        "with tf.device('/cpu:0'):\n",
        "    if ACGAN or MH_LOSS or AM_LOSS:\n",
        "        ac_real_acc=tf.add_n(ac_real_acc)\n",
        "        ac_fake_acc=tf.add_n(ac_fake_acc)\n",
        "        ac_cost=tf.add_n(ac_cost)\n",
        "        if ACGAN:\n",
        "            disc_cost+=ac_cost\n",
        "    else:\n",
        "        ac_cost=tf.zeros([])\n",
        "        ac_real_acc=tf.zeros([])\n",
        "        ac_fake_acc=tf.zeros([]) \n",
        "    if MH_LOSS or AM_LOSS:\n",
        "        disc_cost+= tf.add_n(D_MHINGE_LOSS)\n",
        "        gen_cost=tf.add_n(G_MHINGE_LOSS)+LAMBDA3*tf.add_n(FM_loss)\n",
        "pl_lengths=[]\n",
        "pl_cost=[]\n",
        "if PL_COEFF:\n",
        "    for i in range(NUM_GENERATORS):\n",
        "        with tf.device(DEVICES[i%len(gpu_id)]):\n",
        "            pl_minibatch_shrink=1\n",
        "            pl_decay=0.01\n",
        "            minibatch_size=BATCH_SIZE\n",
        "            pl_minibatch = minibatch_size // pl_minibatch_shrink\n",
        "            pl_latents = noise[i]\n",
        "            pl_labels = fake_labels[i]\n",
        "            fake_images_out, fake_dlatents_out = all_fake_sample[i], dlatents[i] #noise_splits[i//NUM_DEVICES]\n",
        "            num_noise_splits=1\n",
        "            # Compute |J*y|.\n",
        "            pl_noise = tf.random_normal(tf.shape(fake_images_out)) / np.sqrt(HEIGHT*WIDTH)\n",
        "            pl_grads = tf.gradients(tf.reduce_sum(fake_images_out * pl_noise), [fake_dlatents_out])[0]\n",
        "            print('pl_grad:',pl_grads)\n",
        "            pl_lengths=tf.reduce_mean(tf.sqrt(tf.reduce_mean(tf.reduce_sum(tf.square(pl_grads), axis=2), axis=1)))\n",
        "\n",
        "            # Track exponential moving average of |J*y|.\n",
        "            pl_mean_var = lib.get_param('Generator_pl%i'%i, shape=[], initializer=tf.zeros_initializer() ,trainable=False, dtype=tf.float32)\n",
        "            pl_mean = pl_mean_var + pl_decay * (pl_lengths- pl_mean_var)\n",
        "            pl_update = tf.assign(pl_mean_var, pl_mean)\n",
        "            # Calculate (|J*y|-a)^2.\n",
        "            with tf.control_dependencies([pl_update]):\n",
        "                pl_penalty = tf.square(pl_lengths - pl_mean)\n",
        "                pl_cost.append(PL_COEFF*pl_penalty)\n",
        "    with tf.device('/cpu:0'):\n",
        "        gen_cost+=tf.add_n(pl_cost)"
      ],
      "execution_count": 22,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "WARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/tensorflow/python/util/dispatch.py:201: calling argmax (from tensorflow.python.ops.math_ops) with dimension is deprecated and will be removed in a future version.\n",
            "Instructions for updating:\n",
            "Use the `axis` argument instead\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "WARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/tensorflow/python/util/dispatch.py:201: calling argmax (from tensorflow.python.ops.math_ops) with dimension is deprecated and will be removed in a future version.\n",
            "Instructions for updating:\n",
            "Use the `axis` argument instead\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "WARNING:tensorflow:From <ipython-input-22-e3a917e66ceb>:106: to_int32 (from tensorflow.python.ops.math_ops) is deprecated and will be removed in a future version.\n",
            "Instructions for updating:\n",
            "Use `tf.cast` instead.\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "WARNING:tensorflow:From <ipython-input-22-e3a917e66ceb>:106: to_int32 (from tensorflow.python.ops.math_ops) is deprecated and will be removed in a future version.\n",
            "Instructions for updating:\n",
            "Use `tf.cast` instead.\n"
          ],
          "name": "stderr"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "8EFoBy4hz6ty"
      },
      "source": [
        ""
      ],
      "execution_count": 22,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "YchjtRrqJQRM"
      },
      "source": [
        "# for records to be saved in a tsv file\n",
        "\n",
        "num_counter=1\n",
        "record={}\n",
        "names=[]\n",
        "if DATASET in IMAGE_DATASETS and NUM_CHANNELS==3:\n",
        "    name='Inception score'\n",
        "    names.append(name)\n",
        "    record[name]=np.zeros(num_counter)\n",
        "    name='IS_std'\n",
        "    names.append(name)\n",
        "    record[name]=np.zeros(num_counter)\n",
        "    name='FID_TRAIN'\n",
        "    names.append(name)\n",
        "    record[name]=np.zeros(num_counter)\n",
        "    name='FID_VAL'\n",
        "    names.append(name)\n",
        "    record[name]=np.zeros(num_counter)\n",
        "name='D(Xr)-D(Xf)'\n",
        "names.append(name)\n",
        "record[name]=np.zeros(num_counter)\n",
        "name='Wass_dist'\n",
        "names.append(name)\n",
        "record[name]=np.zeros(num_counter)\n",
        "for dst in ['training_set','test_set']:\n",
        "    name='FID_'+dst\n",
        "    names.append(name)\n",
        "    record[name]=np.zeros(num_counter)\n",
        "counter=0"
      ],
      "execution_count": 23,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "JRtoxHhQJQRQ"
      },
      "source": [
        ""
      ],
      "execution_count": 23,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "4TXqBIRJJQRU"
      },
      "source": [
        "ema_params=lib.params_with_name('ema')\n",
        "ema_params=[var for  var in ema_params if 'moving' not in var.name and 'accu' not in var.name]\n",
        "nonema_params=[lib.get_param(var.name.split(':')[0].replace('_ema','')) for var in ema_params]"
      ],
      "execution_count": 24,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "L1wkcOsqJQRW"
      },
      "source": [
        "def warmup():\n",
        "  if ACCU_STATS:\n",
        "        session.run(clear_op)\n",
        "        WARMUP_STEPS=200000//100 if iteration % (10000) == 0 else 100\n",
        "        for i in range(WARMUP_STEPS):\n",
        "            _=session.run(fake_sample)"
      ],
      "execution_count": 25,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "EcqvT_woJQRY"
      },
      "source": [
        "from tensorflow.python.ops import state_ops\n",
        "def ema_update(v, value, decay=0.9999):\n",
        "    #if not v.trainable and '.u_var' not in v.name: decay= 0.\n",
        "    return state_ops.assign_sub(v, (v - value) * (1.-decay))\n",
        "def get_ema_op():\n",
        "    return tf.group(*[ema_update(ema_params[i], nonema_params[i], .9999) for i in range(len(ema_params))]) if EMA else tf.no_op()\n"
      ],
      "execution_count": 26,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "TwVS1hYEJQRb"
      },
      "source": [
        ""
      ],
      "execution_count": 26,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Ik6N4faGJQRe"
      },
      "source": [
        "#from keras_radam.training import RAdamOptimizer\n",
        "def increment_p(p=d_p,offset=4*BATCH_SIZE/5000000.):\n",
        "    return  tf.cond(p<1,functools.partial(state_ops.assign_add,ref=p,value=offset),functools.partial(state_ops.assign,ref=p,value=1.))\n",
        "def decrement_p(p=d_p,offset=4*BATCH_SIZE/5000000.):\n",
        "    return tf.cond(p>0,functools.partial(state_ops.assign_sub,ref=p,value=offset),functools.partial(state_ops.assign,ref=p,value=0.))\n",
        "disc_params = lib.params_with_name('Discriminator')\n",
        "gen_params = lib.params_with_name('Generator')\n",
        "with tf.device('/cpu:0') and session.as_default():\n",
        "    gen_opt = tf.train.AdamOptimizer(learning_rate=lr, beta1=BETA1, beta2=BETA2)\n",
        "    disc_opt = tf.train.AdamOptimizer(learning_rate=lr, beta1=BETA1, beta2=BETA2)\n",
        "    gen_train_op = gen_opt.minimize(gen_cost,var_list=gen_params, colocate_gradients_with_ops=True)\n",
        "    disc_train_op = disc_opt.minimize(disc_cost,var_list=disc_params, colocate_gradients_with_ops=True)\n",
        "    with tf.control_dependencies([disc_train_op]):\n",
        "        with tf.control_dependencies([tf.scatter_update(indicator,d_counter%4,indicator_stat)]):\n",
        "          with tf.control_dependencies([tf.assign_add(d_counter, 1)]):\n",
        "            def update_p_fn():\n",
        "                return tf.cond(tf.reduce_mean(indicator)>ADA_TARGET,increment_p, decrement_p)\n",
        "                \n",
        "            updated_p_op=tf.cond(update_p,update_p_fn,lambda:d_p)\n",
        "            with tf.control_dependencies([updated_p_op]):\n",
        "                    disc_train_op=tf.no_op()\n",
        "    if EMA:\n",
        "        with tf.control_dependencies([gen_train_op]):\n",
        "            gen_train_op = tf.cond(curr_iter> EMA_ITER, get_ema_op, tf.no_op)"
      ],
      "execution_count": 27,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "vCOBKNfIJQRh",
        "outputId": "a369d756-133d-4028-ceeb-100f79174f60",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        }
      },
      "source": [
        "g_params_count=0\n",
        "d_params_count=0\n",
        "print(\"Generator params:\")\n",
        "for var in lib.params_with_name('Generator'):\n",
        "    print(var.device, var.name, var.get_shape(), np.prod(var.get_shape()))\n",
        "    g_params_count+=int(np.prod(var.get_shape()))\n",
        "print(\"Discriminator params:\")\n",
        "for var in lib.params_with_name('Discriminator'):\n",
        "    print(var.device, var.name, var.get_shape(), np.prod(var.get_shape()))\n",
        "    d_params_count+=int(np.prod(var.get_shape()))\n",
        "print(\"G params total:{:,}\".format(int(g_params_count)))\n",
        "print(\"D params total:{:,}\".format(int(d_params_count)))\n",
        "print(\"All params total:{:,}\".format(int(g_params_count+d_params_count)))"
      ],
      "execution_count": 28,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Generator params:\n",
            " Generator_alpha:0 (10,) 10\n",
            "/device:GPU:0 Generator.0.embed.W:0 (10, 128) 1280\n",
            "/device:GPU:0 Generator.0dense.W:0 (20, 4096) 81920\n",
            "/device:GPU:0 Generator.0dense.W.u_var:0 (20, 1) 20\n",
            "/device:GPU:0 Generator.0.up0.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.0.up0.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.0.up0.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.0.up0.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.0.up0.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.0.up0.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.0.up0.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.0.up0.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.0.up0.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.0.up0.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.0.up0.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.0.up0.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.0.up0.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.0.up0.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.0.up0.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator.0.up0.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator.0.up0.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.0.up1.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.0.up1.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.0.up1.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.0.up1.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.0.up1.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.0.up1.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.0.up1.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.0.up1.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.0.up1.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.0.up1.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.0.up1.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.0.up1.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.0.up1.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.0.up1.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.0.up1.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator.0.up1.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator.0.up1.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.0.up2.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.0.up2.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.0.up2.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.0.up2.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.0.up2.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.0.up2.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.0.up2.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.0.up2.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.0.up2.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.0.up2.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.0.up2.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.0.up2.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.0.up2.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.0.up2.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.0.up2.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator.0.up2.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator.0.up2.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.0.last_bn.gamma:0 (256,) 256\n",
            "/device:GPU:0 Generator.0.last_bn.beta:0 (256,) 256\n",
            "/device:GPU:0 Generator.0.last_conv.Filters:0 (3, 3, 256, 3) 6912\n",
            "/device:GPU:0 Generator.0.last_conv.Filters.u_var:0 (1, 3) 3\n",
            "/device:GPU:0 Generator.0.last_conv.Biases:0 (3,) 3\n",
            "/device:GPU:0 Generator_ema.0.embed.W:0 (10, 128) 1280\n",
            "/device:GPU:0 Generator_ema.0dense.W:0 (20, 4096) 81920\n",
            "/device:GPU:0 Generator_ema.0dense.W.u_var:0 (20, 1) 20\n",
            "/device:GPU:0 Generator_ema.0.up0.cbn1.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.0.up0.cbn1.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.0.up0.cbn1.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.0.up0.cbn1.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.0.up0.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.0.up0.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.0.up0.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.0.up0.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.0.up0.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.0.up0.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.0.up0.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.0.up0.cbn2.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.0.up0.cbn2.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.0.up0.cbn2.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.0.up0.cbn2.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.0.up0.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.0.up0.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.0.up0.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.0.up0.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.0.up0.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.0.up0.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.0.up0.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.0.up0.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator_ema.0.up0.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator_ema.0.up0.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.0.up1.cbn1.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.0.up1.cbn1.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.0.up1.cbn1.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.0.up1.cbn1.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.0.up1.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.0.up1.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.0.up1.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.0.up1.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.0.up1.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.0.up1.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.0.up1.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.0.up1.cbn2.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.0.up1.cbn2.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.0.up1.cbn2.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.0.up1.cbn2.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.0.up1.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.0.up1.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.0.up1.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.0.up1.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.0.up1.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.0.up1.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.0.up1.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.0.up1.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator_ema.0.up1.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator_ema.0.up1.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.0.up2.cbn1.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.0.up2.cbn1.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.0.up2.cbn1.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.0.up2.cbn1.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.0.up2.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.0.up2.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.0.up2.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.0.up2.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.0.up2.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.0.up2.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.0.up2.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.0.up2.cbn2.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.0.up2.cbn2.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.0.up2.cbn2.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.0.up2.cbn2.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.0.up2.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.0.up2.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.0.up2.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.0.up2.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.0.up2.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.0.up2.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.0.up2.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.0.up2.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator_ema.0.up2.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator_ema.0.up2.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.0.last_bn.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.0.last_bn.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.0.last_bn.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.0.last_bn.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.0.last_bn.gamma:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.0.last_bn.beta:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.0.last_conv.Filters:0 (3, 3, 256, 3) 6912\n",
            "/device:GPU:0 Generator_ema.0.last_conv.Filters.u_var:0 (1, 3) 3\n",
            "/device:GPU:0 Generator_ema.0.last_conv.Biases:0 (3,) 3\n",
            "/device:GPU:0 Generator.1.embed.W:0 (10, 128) 1280\n",
            "/device:GPU:0 Generator.1dense.W:0 (20, 4096) 81920\n",
            "/device:GPU:0 Generator.1dense.W.u_var:0 (20, 1) 20\n",
            "/device:GPU:0 Generator.1.up0.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.1.up0.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.1.up0.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.1.up0.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.1.up0.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.1.up0.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.1.up0.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.1.up0.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.1.up0.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.1.up0.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.1.up0.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.1.up0.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.1.up0.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.1.up0.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.1.up0.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator.1.up0.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator.1.up0.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.1.up1.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.1.up1.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.1.up1.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.1.up1.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.1.up1.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.1.up1.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.1.up1.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.1.up1.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.1.up1.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.1.up1.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.1.up1.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.1.up1.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.1.up1.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.1.up1.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.1.up1.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator.1.up1.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator.1.up1.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.1.up2.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.1.up2.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.1.up2.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.1.up2.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.1.up2.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.1.up2.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.1.up2.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.1.up2.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.1.up2.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.1.up2.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.1.up2.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.1.up2.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.1.up2.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.1.up2.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.1.up2.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator.1.up2.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator.1.up2.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.1.last_bn.gamma:0 (256,) 256\n",
            "/device:GPU:0 Generator.1.last_bn.beta:0 (256,) 256\n",
            "/device:GPU:0 Generator.1.last_conv.Filters:0 (3, 3, 256, 3) 6912\n",
            "/device:GPU:0 Generator.1.last_conv.Filters.u_var:0 (1, 3) 3\n",
            "/device:GPU:0 Generator.1.last_conv.Biases:0 (3,) 3\n",
            "/device:GPU:0 Generator_ema.1.embed.W:0 (10, 128) 1280\n",
            "/device:GPU:0 Generator_ema.1dense.W:0 (20, 4096) 81920\n",
            "/device:GPU:0 Generator_ema.1dense.W.u_var:0 (20, 1) 20\n",
            "/device:GPU:0 Generator_ema.1.up0.cbn1.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.1.up0.cbn1.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.1.up0.cbn1.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.1.up0.cbn1.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.1.up0.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.1.up0.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.1.up0.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.1.up0.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.1.up0.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.1.up0.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.1.up0.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.1.up0.cbn2.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.1.up0.cbn2.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.1.up0.cbn2.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.1.up0.cbn2.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.1.up0.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.1.up0.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.1.up0.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.1.up0.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.1.up0.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.1.up0.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.1.up0.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.1.up0.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator_ema.1.up0.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator_ema.1.up0.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.1.up1.cbn1.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.1.up1.cbn1.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.1.up1.cbn1.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.1.up1.cbn1.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.1.up1.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.1.up1.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.1.up1.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.1.up1.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.1.up1.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.1.up1.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.1.up1.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.1.up1.cbn2.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.1.up1.cbn2.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.1.up1.cbn2.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.1.up1.cbn2.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.1.up1.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.1.up1.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.1.up1.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.1.up1.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.1.up1.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.1.up1.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.1.up1.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.1.up1.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator_ema.1.up1.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator_ema.1.up1.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.1.up2.cbn1.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.1.up2.cbn1.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.1.up2.cbn1.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.1.up2.cbn1.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.1.up2.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.1.up2.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.1.up2.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.1.up2.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.1.up2.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.1.up2.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.1.up2.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.1.up2.cbn2.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.1.up2.cbn2.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.1.up2.cbn2.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.1.up2.cbn2.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.1.up2.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.1.up2.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.1.up2.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.1.up2.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.1.up2.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.1.up2.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.1.up2.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.1.up2.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator_ema.1.up2.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator_ema.1.up2.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.1.last_bn.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.1.last_bn.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.1.last_bn.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.1.last_bn.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.1.last_bn.gamma:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.1.last_bn.beta:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.1.last_conv.Filters:0 (3, 3, 256, 3) 6912\n",
            "/device:GPU:0 Generator_ema.1.last_conv.Filters.u_var:0 (1, 3) 3\n",
            "/device:GPU:0 Generator_ema.1.last_conv.Biases:0 (3,) 3\n",
            "/device:GPU:0 Generator.2.embed.W:0 (10, 128) 1280\n",
            "/device:GPU:0 Generator.2dense.W:0 (20, 4096) 81920\n",
            "/device:GPU:0 Generator.2dense.W.u_var:0 (20, 1) 20\n",
            "/device:GPU:0 Generator.2.up0.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.2.up0.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.2.up0.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.2.up0.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.2.up0.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.2.up0.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.2.up0.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.2.up0.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.2.up0.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.2.up0.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.2.up0.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.2.up0.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.2.up0.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.2.up0.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.2.up0.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator.2.up0.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator.2.up0.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.2.up1.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.2.up1.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.2.up1.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.2.up1.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.2.up1.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.2.up1.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.2.up1.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.2.up1.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.2.up1.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.2.up1.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.2.up1.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.2.up1.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.2.up1.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.2.up1.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.2.up1.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator.2.up1.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator.2.up1.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.2.up2.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.2.up2.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.2.up2.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.2.up2.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.2.up2.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.2.up2.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.2.up2.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.2.up2.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.2.up2.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.2.up2.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.2.up2.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.2.up2.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.2.up2.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.2.up2.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.2.up2.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator.2.up2.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator.2.up2.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.2.last_bn.gamma:0 (256,) 256\n",
            "/device:GPU:0 Generator.2.last_bn.beta:0 (256,) 256\n",
            "/device:GPU:0 Generator.2.last_conv.Filters:0 (3, 3, 256, 3) 6912\n",
            "/device:GPU:0 Generator.2.last_conv.Filters.u_var:0 (1, 3) 3\n",
            "/device:GPU:0 Generator.2.last_conv.Biases:0 (3,) 3\n",
            "/device:GPU:0 Generator_ema.2.embed.W:0 (10, 128) 1280\n",
            "/device:GPU:0 Generator_ema.2dense.W:0 (20, 4096) 81920\n",
            "/device:GPU:0 Generator_ema.2dense.W.u_var:0 (20, 1) 20\n",
            "/device:GPU:0 Generator_ema.2.up0.cbn1.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.2.up0.cbn1.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.2.up0.cbn1.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.2.up0.cbn1.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.2.up0.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.2.up0.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.2.up0.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.2.up0.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.2.up0.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.2.up0.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.2.up0.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.2.up0.cbn2.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.2.up0.cbn2.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.2.up0.cbn2.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.2.up0.cbn2.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.2.up0.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.2.up0.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.2.up0.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.2.up0.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.2.up0.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.2.up0.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.2.up0.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.2.up0.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator_ema.2.up0.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator_ema.2.up0.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.2.up1.cbn1.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.2.up1.cbn1.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.2.up1.cbn1.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.2.up1.cbn1.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.2.up1.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.2.up1.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.2.up1.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.2.up1.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.2.up1.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.2.up1.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.2.up1.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.2.up1.cbn2.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.2.up1.cbn2.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.2.up1.cbn2.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.2.up1.cbn2.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.2.up1.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.2.up1.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.2.up1.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.2.up1.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.2.up1.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.2.up1.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.2.up1.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.2.up1.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator_ema.2.up1.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator_ema.2.up1.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.2.up2.cbn1.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.2.up2.cbn1.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.2.up2.cbn1.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.2.up2.cbn1.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.2.up2.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.2.up2.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.2.up2.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.2.up2.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.2.up2.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.2.up2.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.2.up2.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.2.up2.cbn2.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.2.up2.cbn2.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.2.up2.cbn2.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.2.up2.cbn2.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.2.up2.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.2.up2.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.2.up2.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.2.up2.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.2.up2.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.2.up2.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.2.up2.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.2.up2.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator_ema.2.up2.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator_ema.2.up2.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.2.last_bn.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.2.last_bn.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.2.last_bn.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.2.last_bn.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.2.last_bn.gamma:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.2.last_bn.beta:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.2.last_conv.Filters:0 (3, 3, 256, 3) 6912\n",
            "/device:GPU:0 Generator_ema.2.last_conv.Filters.u_var:0 (1, 3) 3\n",
            "/device:GPU:0 Generator_ema.2.last_conv.Biases:0 (3,) 3\n",
            "/device:GPU:0 Generator.3.embed.W:0 (10, 128) 1280\n",
            "/device:GPU:0 Generator.3dense.W:0 (20, 4096) 81920\n",
            "/device:GPU:0 Generator.3dense.W.u_var:0 (20, 1) 20\n",
            "/device:GPU:0 Generator.3.up0.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.3.up0.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.3.up0.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.3.up0.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.3.up0.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.3.up0.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.3.up0.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.3.up0.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.3.up0.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.3.up0.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.3.up0.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.3.up0.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.3.up0.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.3.up0.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.3.up0.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator.3.up0.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator.3.up0.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.3.up1.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.3.up1.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.3.up1.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.3.up1.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.3.up1.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.3.up1.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.3.up1.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.3.up1.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.3.up1.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.3.up1.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.3.up1.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.3.up1.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.3.up1.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.3.up1.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.3.up1.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator.3.up1.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator.3.up1.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.3.up2.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.3.up2.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.3.up2.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.3.up2.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.3.up2.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.3.up2.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.3.up2.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.3.up2.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.3.up2.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.3.up2.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.3.up2.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.3.up2.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.3.up2.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.3.up2.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.3.up2.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator.3.up2.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator.3.up2.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.3.last_bn.gamma:0 (256,) 256\n",
            "/device:GPU:0 Generator.3.last_bn.beta:0 (256,) 256\n",
            "/device:GPU:0 Generator.3.last_conv.Filters:0 (3, 3, 256, 3) 6912\n",
            "/device:GPU:0 Generator.3.last_conv.Filters.u_var:0 (1, 3) 3\n",
            "/device:GPU:0 Generator.3.last_conv.Biases:0 (3,) 3\n",
            "/device:GPU:0 Generator_ema.3.embed.W:0 (10, 128) 1280\n",
            "/device:GPU:0 Generator_ema.3dense.W:0 (20, 4096) 81920\n",
            "/device:GPU:0 Generator_ema.3dense.W.u_var:0 (20, 1) 20\n",
            "/device:GPU:0 Generator_ema.3.up0.cbn1.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.3.up0.cbn1.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.3.up0.cbn1.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.3.up0.cbn1.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.3.up0.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.3.up0.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.3.up0.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.3.up0.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.3.up0.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.3.up0.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.3.up0.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.3.up0.cbn2.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.3.up0.cbn2.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.3.up0.cbn2.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.3.up0.cbn2.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.3.up0.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.3.up0.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.3.up0.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.3.up0.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.3.up0.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.3.up0.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.3.up0.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.3.up0.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator_ema.3.up0.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator_ema.3.up0.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.3.up1.cbn1.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.3.up1.cbn1.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.3.up1.cbn1.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.3.up1.cbn1.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.3.up1.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.3.up1.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.3.up1.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.3.up1.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.3.up1.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.3.up1.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.3.up1.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.3.up1.cbn2.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.3.up1.cbn2.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.3.up1.cbn2.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.3.up1.cbn2.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.3.up1.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.3.up1.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.3.up1.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.3.up1.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.3.up1.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.3.up1.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.3.up1.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.3.up1.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator_ema.3.up1.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator_ema.3.up1.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.3.up2.cbn1.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.3.up2.cbn1.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.3.up2.cbn1.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.3.up2.cbn1.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.3.up2.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.3.up2.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.3.up2.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.3.up2.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.3.up2.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.3.up2.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.3.up2.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.3.up2.cbn2.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.3.up2.cbn2.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.3.up2.cbn2.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.3.up2.cbn2.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.3.up2.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.3.up2.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.3.up2.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.3.up2.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.3.up2.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.3.up2.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.3.up2.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.3.up2.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator_ema.3.up2.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator_ema.3.up2.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.3.last_bn.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.3.last_bn.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.3.last_bn.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.3.last_bn.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.3.last_bn.gamma:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.3.last_bn.beta:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.3.last_conv.Filters:0 (3, 3, 256, 3) 6912\n",
            "/device:GPU:0 Generator_ema.3.last_conv.Filters.u_var:0 (1, 3) 3\n",
            "/device:GPU:0 Generator_ema.3.last_conv.Biases:0 (3,) 3\n",
            "/device:GPU:0 Generator.4.embed.W:0 (10, 128) 1280\n",
            "/device:GPU:0 Generator.4dense.W:0 (20, 4096) 81920\n",
            "/device:GPU:0 Generator.4dense.W.u_var:0 (20, 1) 20\n",
            "/device:GPU:0 Generator.4.up0.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.4.up0.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.4.up0.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.4.up0.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.4.up0.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.4.up0.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.4.up0.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.4.up0.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.4.up0.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.4.up0.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.4.up0.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.4.up0.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.4.up0.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.4.up0.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.4.up0.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator.4.up0.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator.4.up0.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.4.up1.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.4.up1.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.4.up1.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.4.up1.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.4.up1.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.4.up1.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.4.up1.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.4.up1.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.4.up1.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.4.up1.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.4.up1.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.4.up1.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.4.up1.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.4.up1.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.4.up1.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator.4.up1.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator.4.up1.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.4.up2.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.4.up2.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.4.up2.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.4.up2.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.4.up2.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.4.up2.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.4.up2.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.4.up2.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.4.up2.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.4.up2.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.4.up2.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.4.up2.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.4.up2.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.4.up2.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.4.up2.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator.4.up2.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator.4.up2.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.4.last_bn.gamma:0 (256,) 256\n",
            "/device:GPU:0 Generator.4.last_bn.beta:0 (256,) 256\n",
            "/device:GPU:0 Generator.4.last_conv.Filters:0 (3, 3, 256, 3) 6912\n",
            "/device:GPU:0 Generator.4.last_conv.Filters.u_var:0 (1, 3) 3\n",
            "/device:GPU:0 Generator.4.last_conv.Biases:0 (3,) 3\n",
            "/device:GPU:0 Generator_ema.4.embed.W:0 (10, 128) 1280\n",
            "/device:GPU:0 Generator_ema.4dense.W:0 (20, 4096) 81920\n",
            "/device:GPU:0 Generator_ema.4dense.W.u_var:0 (20, 1) 20\n",
            "/device:GPU:0 Generator_ema.4.up0.cbn1.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.4.up0.cbn1.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.4.up0.cbn1.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.4.up0.cbn1.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.4.up0.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.4.up0.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.4.up0.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.4.up0.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.4.up0.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.4.up0.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.4.up0.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.4.up0.cbn2.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.4.up0.cbn2.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.4.up0.cbn2.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.4.up0.cbn2.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.4.up0.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.4.up0.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.4.up0.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.4.up0.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.4.up0.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.4.up0.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.4.up0.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.4.up0.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator_ema.4.up0.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator_ema.4.up0.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.4.up1.cbn1.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.4.up1.cbn1.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.4.up1.cbn1.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.4.up1.cbn1.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.4.up1.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.4.up1.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.4.up1.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.4.up1.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.4.up1.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.4.up1.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.4.up1.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.4.up1.cbn2.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.4.up1.cbn2.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.4.up1.cbn2.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.4.up1.cbn2.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.4.up1.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.4.up1.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.4.up1.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.4.up1.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.4.up1.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.4.up1.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.4.up1.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.4.up1.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator_ema.4.up1.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator_ema.4.up1.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.4.up2.cbn1.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.4.up2.cbn1.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.4.up2.cbn1.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.4.up2.cbn1.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.4.up2.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.4.up2.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.4.up2.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.4.up2.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.4.up2.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.4.up2.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.4.up2.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.4.up2.cbn2.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.4.up2.cbn2.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.4.up2.cbn2.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.4.up2.cbn2.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.4.up2.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.4.up2.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.4.up2.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.4.up2.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.4.up2.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.4.up2.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.4.up2.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.4.up2.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator_ema.4.up2.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator_ema.4.up2.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.4.last_bn.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.4.last_bn.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.4.last_bn.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.4.last_bn.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.4.last_bn.gamma:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.4.last_bn.beta:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.4.last_conv.Filters:0 (3, 3, 256, 3) 6912\n",
            "/device:GPU:0 Generator_ema.4.last_conv.Filters.u_var:0 (1, 3) 3\n",
            "/device:GPU:0 Generator_ema.4.last_conv.Biases:0 (3,) 3\n",
            "/device:GPU:0 Generator.5.embed.W:0 (10, 128) 1280\n",
            "/device:GPU:0 Generator.5dense.W:0 (20, 4096) 81920\n",
            "/device:GPU:0 Generator.5dense.W.u_var:0 (20, 1) 20\n",
            "/device:GPU:0 Generator.5.up0.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.5.up0.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.5.up0.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.5.up0.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.5.up0.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.5.up0.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.5.up0.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.5.up0.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.5.up0.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.5.up0.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.5.up0.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.5.up0.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.5.up0.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.5.up0.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.5.up0.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator.5.up0.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator.5.up0.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.5.up1.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.5.up1.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.5.up1.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.5.up1.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.5.up1.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.5.up1.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.5.up1.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.5.up1.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.5.up1.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.5.up1.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.5.up1.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.5.up1.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.5.up1.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.5.up1.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.5.up1.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator.5.up1.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator.5.up1.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.5.up2.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.5.up2.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.5.up2.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.5.up2.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.5.up2.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.5.up2.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.5.up2.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.5.up2.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.5.up2.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.5.up2.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.5.up2.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.5.up2.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.5.up2.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.5.up2.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.5.up2.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator.5.up2.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator.5.up2.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.5.last_bn.gamma:0 (256,) 256\n",
            "/device:GPU:0 Generator.5.last_bn.beta:0 (256,) 256\n",
            "/device:GPU:0 Generator.5.last_conv.Filters:0 (3, 3, 256, 3) 6912\n",
            "/device:GPU:0 Generator.5.last_conv.Filters.u_var:0 (1, 3) 3\n",
            "/device:GPU:0 Generator.5.last_conv.Biases:0 (3,) 3\n",
            "/device:GPU:0 Generator_ema.5.embed.W:0 (10, 128) 1280\n",
            "/device:GPU:0 Generator_ema.5dense.W:0 (20, 4096) 81920\n",
            "/device:GPU:0 Generator_ema.5dense.W.u_var:0 (20, 1) 20\n",
            "/device:GPU:0 Generator_ema.5.up0.cbn1.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.5.up0.cbn1.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.5.up0.cbn1.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.5.up0.cbn1.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.5.up0.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.5.up0.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.5.up0.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.5.up0.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.5.up0.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.5.up0.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.5.up0.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.5.up0.cbn2.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.5.up0.cbn2.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.5.up0.cbn2.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.5.up0.cbn2.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.5.up0.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.5.up0.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.5.up0.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.5.up0.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.5.up0.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.5.up0.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.5.up0.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.5.up0.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator_ema.5.up0.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator_ema.5.up0.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.5.up1.cbn1.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.5.up1.cbn1.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.5.up1.cbn1.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.5.up1.cbn1.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.5.up1.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.5.up1.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.5.up1.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.5.up1.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.5.up1.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.5.up1.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.5.up1.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.5.up1.cbn2.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.5.up1.cbn2.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.5.up1.cbn2.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.5.up1.cbn2.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.5.up1.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.5.up1.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.5.up1.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.5.up1.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.5.up1.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.5.up1.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.5.up1.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.5.up1.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator_ema.5.up1.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator_ema.5.up1.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.5.up2.cbn1.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.5.up2.cbn1.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.5.up2.cbn1.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.5.up2.cbn1.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.5.up2.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.5.up2.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.5.up2.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.5.up2.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.5.up2.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.5.up2.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.5.up2.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.5.up2.cbn2.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.5.up2.cbn2.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.5.up2.cbn2.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.5.up2.cbn2.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.5.up2.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.5.up2.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.5.up2.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.5.up2.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.5.up2.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.5.up2.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.5.up2.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.5.up2.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator_ema.5.up2.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator_ema.5.up2.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.5.last_bn.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.5.last_bn.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.5.last_bn.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.5.last_bn.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.5.last_bn.gamma:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.5.last_bn.beta:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.5.last_conv.Filters:0 (3, 3, 256, 3) 6912\n",
            "/device:GPU:0 Generator_ema.5.last_conv.Filters.u_var:0 (1, 3) 3\n",
            "/device:GPU:0 Generator_ema.5.last_conv.Biases:0 (3,) 3\n",
            "/device:GPU:0 Generator.6.embed.W:0 (10, 128) 1280\n",
            "/device:GPU:0 Generator.6dense.W:0 (20, 4096) 81920\n",
            "/device:GPU:0 Generator.6dense.W.u_var:0 (20, 1) 20\n",
            "/device:GPU:0 Generator.6.up0.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.6.up0.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.6.up0.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.6.up0.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.6.up0.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.6.up0.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.6.up0.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.6.up0.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.6.up0.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.6.up0.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.6.up0.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.6.up0.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.6.up0.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.6.up0.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.6.up0.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator.6.up0.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator.6.up0.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.6.up1.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.6.up1.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.6.up1.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.6.up1.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.6.up1.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.6.up1.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.6.up1.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.6.up1.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.6.up1.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.6.up1.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.6.up1.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.6.up1.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.6.up1.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.6.up1.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.6.up1.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator.6.up1.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator.6.up1.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.6.up2.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.6.up2.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.6.up2.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.6.up2.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.6.up2.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.6.up2.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.6.up2.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.6.up2.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.6.up2.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.6.up2.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.6.up2.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.6.up2.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.6.up2.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.6.up2.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.6.up2.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator.6.up2.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator.6.up2.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.6.last_bn.gamma:0 (256,) 256\n",
            "/device:GPU:0 Generator.6.last_bn.beta:0 (256,) 256\n",
            "/device:GPU:0 Generator.6.last_conv.Filters:0 (3, 3, 256, 3) 6912\n",
            "/device:GPU:0 Generator.6.last_conv.Filters.u_var:0 (1, 3) 3\n",
            "/device:GPU:0 Generator.6.last_conv.Biases:0 (3,) 3\n",
            "/device:GPU:0 Generator_ema.6.embed.W:0 (10, 128) 1280\n",
            "/device:GPU:0 Generator_ema.6dense.W:0 (20, 4096) 81920\n",
            "/device:GPU:0 Generator_ema.6dense.W.u_var:0 (20, 1) 20\n",
            "/device:GPU:0 Generator_ema.6.up0.cbn1.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.6.up0.cbn1.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.6.up0.cbn1.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.6.up0.cbn1.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.6.up0.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.6.up0.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.6.up0.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.6.up0.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.6.up0.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.6.up0.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.6.up0.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.6.up0.cbn2.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.6.up0.cbn2.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.6.up0.cbn2.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.6.up0.cbn2.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.6.up0.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.6.up0.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.6.up0.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.6.up0.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.6.up0.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.6.up0.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.6.up0.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.6.up0.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator_ema.6.up0.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator_ema.6.up0.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.6.up1.cbn1.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.6.up1.cbn1.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.6.up1.cbn1.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.6.up1.cbn1.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.6.up1.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.6.up1.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.6.up1.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.6.up1.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.6.up1.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.6.up1.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.6.up1.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.6.up1.cbn2.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.6.up1.cbn2.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.6.up1.cbn2.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.6.up1.cbn2.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.6.up1.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.6.up1.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.6.up1.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.6.up1.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.6.up1.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.6.up1.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.6.up1.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.6.up1.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator_ema.6.up1.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator_ema.6.up1.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.6.up2.cbn1.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.6.up2.cbn1.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.6.up2.cbn1.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.6.up2.cbn1.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.6.up2.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.6.up2.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.6.up2.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.6.up2.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.6.up2.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.6.up2.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.6.up2.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.6.up2.cbn2.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.6.up2.cbn2.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.6.up2.cbn2.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.6.up2.cbn2.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.6.up2.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.6.up2.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.6.up2.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.6.up2.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.6.up2.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.6.up2.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.6.up2.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.6.up2.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator_ema.6.up2.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator_ema.6.up2.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.6.last_bn.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.6.last_bn.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.6.last_bn.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.6.last_bn.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.6.last_bn.gamma:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.6.last_bn.beta:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.6.last_conv.Filters:0 (3, 3, 256, 3) 6912\n",
            "/device:GPU:0 Generator_ema.6.last_conv.Filters.u_var:0 (1, 3) 3\n",
            "/device:GPU:0 Generator_ema.6.last_conv.Biases:0 (3,) 3\n",
            "/device:GPU:0 Generator.7.embed.W:0 (10, 128) 1280\n",
            "/device:GPU:0 Generator.7dense.W:0 (20, 4096) 81920\n",
            "/device:GPU:0 Generator.7dense.W.u_var:0 (20, 1) 20\n",
            "/device:GPU:0 Generator.7.up0.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.7.up0.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.7.up0.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.7.up0.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.7.up0.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.7.up0.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.7.up0.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.7.up0.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.7.up0.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.7.up0.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.7.up0.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.7.up0.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.7.up0.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.7.up0.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.7.up0.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator.7.up0.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator.7.up0.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.7.up1.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.7.up1.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.7.up1.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.7.up1.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.7.up1.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.7.up1.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.7.up1.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.7.up1.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.7.up1.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.7.up1.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.7.up1.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.7.up1.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.7.up1.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.7.up1.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.7.up1.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator.7.up1.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator.7.up1.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.7.up2.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.7.up2.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.7.up2.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.7.up2.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.7.up2.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.7.up2.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.7.up2.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.7.up2.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.7.up2.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.7.up2.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.7.up2.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.7.up2.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.7.up2.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.7.up2.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.7.up2.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator.7.up2.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator.7.up2.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.7.last_bn.gamma:0 (256,) 256\n",
            "/device:GPU:0 Generator.7.last_bn.beta:0 (256,) 256\n",
            "/device:GPU:0 Generator.7.last_conv.Filters:0 (3, 3, 256, 3) 6912\n",
            "/device:GPU:0 Generator.7.last_conv.Filters.u_var:0 (1, 3) 3\n",
            "/device:GPU:0 Generator.7.last_conv.Biases:0 (3,) 3\n",
            "/device:GPU:0 Generator_ema.7.embed.W:0 (10, 128) 1280\n",
            "/device:GPU:0 Generator_ema.7dense.W:0 (20, 4096) 81920\n",
            "/device:GPU:0 Generator_ema.7dense.W.u_var:0 (20, 1) 20\n",
            "/device:GPU:0 Generator_ema.7.up0.cbn1.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.7.up0.cbn1.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.7.up0.cbn1.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.7.up0.cbn1.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.7.up0.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.7.up0.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.7.up0.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.7.up0.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.7.up0.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.7.up0.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.7.up0.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.7.up0.cbn2.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.7.up0.cbn2.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.7.up0.cbn2.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.7.up0.cbn2.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.7.up0.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.7.up0.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.7.up0.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.7.up0.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.7.up0.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.7.up0.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.7.up0.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.7.up0.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator_ema.7.up0.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator_ema.7.up0.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.7.up1.cbn1.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.7.up1.cbn1.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.7.up1.cbn1.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.7.up1.cbn1.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.7.up1.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.7.up1.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.7.up1.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.7.up1.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.7.up1.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.7.up1.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.7.up1.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.7.up1.cbn2.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.7.up1.cbn2.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.7.up1.cbn2.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.7.up1.cbn2.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.7.up1.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.7.up1.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.7.up1.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.7.up1.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.7.up1.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.7.up1.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.7.up1.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.7.up1.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator_ema.7.up1.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator_ema.7.up1.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.7.up2.cbn1.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.7.up2.cbn1.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.7.up2.cbn1.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.7.up2.cbn1.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.7.up2.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.7.up2.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.7.up2.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.7.up2.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.7.up2.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.7.up2.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.7.up2.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.7.up2.cbn2.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.7.up2.cbn2.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.7.up2.cbn2.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.7.up2.cbn2.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.7.up2.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.7.up2.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.7.up2.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.7.up2.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.7.up2.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.7.up2.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.7.up2.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.7.up2.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator_ema.7.up2.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator_ema.7.up2.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.7.last_bn.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.7.last_bn.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.7.last_bn.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.7.last_bn.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.7.last_bn.gamma:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.7.last_bn.beta:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.7.last_conv.Filters:0 (3, 3, 256, 3) 6912\n",
            "/device:GPU:0 Generator_ema.7.last_conv.Filters.u_var:0 (1, 3) 3\n",
            "/device:GPU:0 Generator_ema.7.last_conv.Biases:0 (3,) 3\n",
            "/device:GPU:0 Generator.8.embed.W:0 (10, 128) 1280\n",
            "/device:GPU:0 Generator.8dense.W:0 (20, 4096) 81920\n",
            "/device:GPU:0 Generator.8dense.W.u_var:0 (20, 1) 20\n",
            "/device:GPU:0 Generator.8.up0.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.8.up0.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.8.up0.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.8.up0.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.8.up0.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.8.up0.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.8.up0.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.8.up0.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.8.up0.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.8.up0.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.8.up0.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.8.up0.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.8.up0.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.8.up0.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.8.up0.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator.8.up0.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator.8.up0.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.8.up1.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.8.up1.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.8.up1.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.8.up1.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.8.up1.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.8.up1.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.8.up1.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.8.up1.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.8.up1.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.8.up1.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.8.up1.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.8.up1.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.8.up1.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.8.up1.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.8.up1.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator.8.up1.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator.8.up1.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.8.up2.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.8.up2.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.8.up2.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.8.up2.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.8.up2.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.8.up2.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.8.up2.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.8.up2.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.8.up2.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.8.up2.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.8.up2.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.8.up2.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.8.up2.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.8.up2.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.8.up2.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator.8.up2.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator.8.up2.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.8.last_bn.gamma:0 (256,) 256\n",
            "/device:GPU:0 Generator.8.last_bn.beta:0 (256,) 256\n",
            "/device:GPU:0 Generator.8.last_conv.Filters:0 (3, 3, 256, 3) 6912\n",
            "/device:GPU:0 Generator.8.last_conv.Filters.u_var:0 (1, 3) 3\n",
            "/device:GPU:0 Generator.8.last_conv.Biases:0 (3,) 3\n",
            "/device:GPU:0 Generator_ema.8.embed.W:0 (10, 128) 1280\n",
            "/device:GPU:0 Generator_ema.8dense.W:0 (20, 4096) 81920\n",
            "/device:GPU:0 Generator_ema.8dense.W.u_var:0 (20, 1) 20\n",
            "/device:GPU:0 Generator_ema.8.up0.cbn1.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.8.up0.cbn1.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.8.up0.cbn1.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.8.up0.cbn1.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.8.up0.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.8.up0.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.8.up0.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.8.up0.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.8.up0.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.8.up0.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.8.up0.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.8.up0.cbn2.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.8.up0.cbn2.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.8.up0.cbn2.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.8.up0.cbn2.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.8.up0.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.8.up0.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.8.up0.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.8.up0.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.8.up0.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.8.up0.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.8.up0.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.8.up0.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator_ema.8.up0.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator_ema.8.up0.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.8.up1.cbn1.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.8.up1.cbn1.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.8.up1.cbn1.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.8.up1.cbn1.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.8.up1.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.8.up1.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.8.up1.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.8.up1.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.8.up1.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.8.up1.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.8.up1.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.8.up1.cbn2.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.8.up1.cbn2.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.8.up1.cbn2.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.8.up1.cbn2.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.8.up1.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.8.up1.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.8.up1.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.8.up1.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.8.up1.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.8.up1.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.8.up1.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.8.up1.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator_ema.8.up1.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator_ema.8.up1.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.8.up2.cbn1.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.8.up2.cbn1.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.8.up2.cbn1.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.8.up2.cbn1.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.8.up2.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.8.up2.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.8.up2.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.8.up2.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.8.up2.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.8.up2.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.8.up2.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.8.up2.cbn2.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.8.up2.cbn2.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.8.up2.cbn2.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.8.up2.cbn2.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.8.up2.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.8.up2.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.8.up2.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.8.up2.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.8.up2.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.8.up2.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.8.up2.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.8.up2.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator_ema.8.up2.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator_ema.8.up2.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.8.last_bn.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.8.last_bn.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.8.last_bn.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.8.last_bn.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.8.last_bn.gamma:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.8.last_bn.beta:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.8.last_conv.Filters:0 (3, 3, 256, 3) 6912\n",
            "/device:GPU:0 Generator_ema.8.last_conv.Filters.u_var:0 (1, 3) 3\n",
            "/device:GPU:0 Generator_ema.8.last_conv.Biases:0 (3,) 3\n",
            "/device:GPU:0 Generator.9.embed.W:0 (10, 128) 1280\n",
            "/device:GPU:0 Generator.9dense.W:0 (20, 4096) 81920\n",
            "/device:GPU:0 Generator.9dense.W.u_var:0 (20, 1) 20\n",
            "/device:GPU:0 Generator.9.up0.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.9.up0.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.9.up0.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.9.up0.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.9.up0.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.9.up0.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.9.up0.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.9.up0.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.9.up0.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.9.up0.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.9.up0.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.9.up0.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.9.up0.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.9.up0.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.9.up0.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator.9.up0.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator.9.up0.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.9.up1.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.9.up1.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.9.up1.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.9.up1.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.9.up1.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.9.up1.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.9.up1.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.9.up1.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.9.up1.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.9.up1.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.9.up1.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.9.up1.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.9.up1.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.9.up1.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.9.up1.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator.9.up1.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator.9.up1.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.9.up2.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.9.up2.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.9.up2.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.9.up2.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.9.up2.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.9.up2.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.9.up2.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.9.up2.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.9.up2.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.9.up2.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator.9.up2.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator.9.up2.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator.9.up2.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator.9.up2.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.9.up2.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator.9.up2.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator.9.up2.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator.9.last_bn.gamma:0 (256,) 256\n",
            "/device:GPU:0 Generator.9.last_bn.beta:0 (256,) 256\n",
            "/device:GPU:0 Generator.9.last_conv.Filters:0 (3, 3, 256, 3) 6912\n",
            "/device:GPU:0 Generator.9.last_conv.Filters.u_var:0 (1, 3) 3\n",
            "/device:GPU:0 Generator.9.last_conv.Biases:0 (3,) 3\n",
            "/device:GPU:0 Generator_ema.9.embed.W:0 (10, 128) 1280\n",
            "/device:GPU:0 Generator_ema.9dense.W:0 (20, 4096) 81920\n",
            "/device:GPU:0 Generator_ema.9dense.W.u_var:0 (20, 1) 20\n",
            "/device:GPU:0 Generator_ema.9.up0.cbn1.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.9.up0.cbn1.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.9.up0.cbn1.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.9.up0.cbn1.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.9.up0.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.9.up0.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.9.up0.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.9.up0.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.9.up0.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.9.up0.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.9.up0.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.9.up0.cbn2.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.9.up0.cbn2.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.9.up0.cbn2.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.9.up0.cbn2.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.9.up0.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.9.up0.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.9.up0.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.9.up0.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.9.up0.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.9.up0.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.9.up0.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.9.up0.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator_ema.9.up0.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator_ema.9.up0.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.9.up1.cbn1.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.9.up1.cbn1.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.9.up1.cbn1.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.9.up1.cbn1.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.9.up1.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.9.up1.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.9.up1.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.9.up1.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.9.up1.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.9.up1.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.9.up1.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.9.up1.cbn2.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.9.up1.cbn2.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.9.up1.cbn2.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.9.up1.cbn2.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.9.up1.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.9.up1.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.9.up1.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.9.up1.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.9.up1.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.9.up1.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.9.up1.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.9.up1.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator_ema.9.up1.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator_ema.9.up1.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.9.up2.cbn1.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.9.up2.cbn1.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.9.up2.cbn1.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.9.up2.cbn1.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.9.up2.cbn1.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.9.up2.cbn1.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.9.up2.cbn1.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.9.up2.cbn1.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.9.up2.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.9.up2.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.9.up2.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.9.up2.cbn2.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.9.up2.cbn2.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.9.up2.cbn2.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.9.up2.cbn2.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.9.up2.cbn2.Gamma.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.9.up2.cbn2.Gamma.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.9.up2.cbn2.Beta.W:0 (148, 256) 37888\n",
            "/device:GPU:0 Generator_ema.9.up2.cbn2.Beta.W.u_var:0 (148, 1) 148\n",
            "/device:GPU:0 Generator_ema.9.up2.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Generator_ema.9.up2.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Generator_ema.9.up2.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.9.up2.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Generator_ema.9.up2.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Generator_ema.9.up2.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.9.last_bn.accu_mean:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.9.last_bn.accu_variance:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.9.last_bn.accu_counter:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.9.last_bn.update_accus:0 () 1.0\n",
            "/device:GPU:0 Generator_ema.9.last_bn.gamma:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.9.last_bn.beta:0 (256,) 256\n",
            "/device:GPU:0 Generator_ema.9.last_conv.Filters:0 (3, 3, 256, 3) 6912\n",
            "/device:GPU:0 Generator_ema.9.last_conv.Filters.u_var:0 (1, 3) 3\n",
            "/device:GPU:0 Generator_ema.9.last_conv.Biases:0 (3,) 3\n",
            "Discriminator params:\n",
            " Discriminator_alpha:0 (10,) 10\n",
            " Discriminator_d_counter:0 () 1.0\n",
            " Discriminator_indicator:0 (4,) 4\n",
            " Discriminator_p:0 () 1.0\n",
            "/device:GPU:0 Discriminator.0.resdown1.conv1.Filters:0 (3, 3, 3, 256) 6912\n",
            "/device:GPU:0 Discriminator.0.resdown1.conv1.Filters.u_var:0 (27, 1) 27\n",
            "/device:GPU:0 Discriminator.0.resdown1.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.0.resdown1.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.0.resdown1.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.0.resdown1.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.0.resdown1.conv_sc.Filters:0 (1, 1, 3, 256) 768\n",
            "/device:GPU:0 Discriminator.0.resdown1.conv_sc.Filters.u_var:0 (3, 1) 3\n",
            "/device:GPU:0 Discriminator.0.resdown1.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.0.resdown2.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.0.resdown2.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.0.resdown2.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.0.resdown2.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.0.resdown2.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.0.resdown2.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.0.resdown2.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Discriminator.0.resdown2.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Discriminator.0.resdown2.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.0.res8.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.0.res8.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.0.res8.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.0.res8.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.0.res8.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.0.res8.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.0.res16.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.0.res16.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.0.res16.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.0.res16.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.0.res16.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.0.res16.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.0.embed.W:0 (10, 256) 2560\n",
            "/device:GPU:0 Discriminator.0.embed.W.u_var:0 (10, 1) 10\n",
            "/device:GPU:0 Discriminator0.Output.W:0 (256, 1) 256\n",
            "/device:GPU:0 Discriminator0.Output.W.u_var:0 (1, 1) 1\n",
            "/device:GPU:0 Discriminator0.Output.b:0 (1,) 1\n",
            "/device:GPU:0 Discriminator.0.MHINGEGANOutput.W:0 (256, 11) 2816\n",
            "/device:GPU:0 Discriminator.0.MHINGEGANOutput.W.u_var:0 (1, 11) 11\n",
            "/device:GPU:0 Discriminator.0.MHINGEGANOutput.b:0 (11,) 11\n",
            "/device:GPU:0 Discriminator.1.resdown1.conv1.Filters:0 (3, 3, 3, 256) 6912\n",
            "/device:GPU:0 Discriminator.1.resdown1.conv1.Filters.u_var:0 (27, 1) 27\n",
            "/device:GPU:0 Discriminator.1.resdown1.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.1.resdown1.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.1.resdown1.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.1.resdown1.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.1.resdown1.conv_sc.Filters:0 (1, 1, 3, 256) 768\n",
            "/device:GPU:0 Discriminator.1.resdown1.conv_sc.Filters.u_var:0 (3, 1) 3\n",
            "/device:GPU:0 Discriminator.1.resdown1.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.1.resdown2.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.1.resdown2.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.1.resdown2.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.1.resdown2.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.1.resdown2.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.1.resdown2.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.1.resdown2.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Discriminator.1.resdown2.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Discriminator.1.resdown2.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.1.res8.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.1.res8.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.1.res8.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.1.res8.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.1.res8.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.1.res8.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.1.res16.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.1.res16.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.1.res16.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.1.res16.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.1.res16.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.1.res16.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.1.embed.W:0 (10, 256) 2560\n",
            "/device:GPU:0 Discriminator.1.embed.W.u_var:0 (10, 1) 10\n",
            "/device:GPU:0 Discriminator1.Output.W:0 (256, 1) 256\n",
            "/device:GPU:0 Discriminator1.Output.W.u_var:0 (1, 1) 1\n",
            "/device:GPU:0 Discriminator1.Output.b:0 (1,) 1\n",
            "/device:GPU:0 Discriminator.1.MHINGEGANOutput.W:0 (256, 11) 2816\n",
            "/device:GPU:0 Discriminator.1.MHINGEGANOutput.W.u_var:0 (1, 11) 11\n",
            "/device:GPU:0 Discriminator.1.MHINGEGANOutput.b:0 (11,) 11\n",
            "/device:GPU:0 Discriminator.2.resdown1.conv1.Filters:0 (3, 3, 3, 256) 6912\n",
            "/device:GPU:0 Discriminator.2.resdown1.conv1.Filters.u_var:0 (27, 1) 27\n",
            "/device:GPU:0 Discriminator.2.resdown1.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.2.resdown1.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.2.resdown1.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.2.resdown1.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.2.resdown1.conv_sc.Filters:0 (1, 1, 3, 256) 768\n",
            "/device:GPU:0 Discriminator.2.resdown1.conv_sc.Filters.u_var:0 (3, 1) 3\n",
            "/device:GPU:0 Discriminator.2.resdown1.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.2.resdown2.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.2.resdown2.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.2.resdown2.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.2.resdown2.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.2.resdown2.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.2.resdown2.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.2.resdown2.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Discriminator.2.resdown2.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Discriminator.2.resdown2.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.2.res8.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.2.res8.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.2.res8.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.2.res8.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.2.res8.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.2.res8.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.2.res16.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.2.res16.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.2.res16.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.2.res16.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.2.res16.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.2.res16.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.2.embed.W:0 (10, 256) 2560\n",
            "/device:GPU:0 Discriminator.2.embed.W.u_var:0 (10, 1) 10\n",
            "/device:GPU:0 Discriminator2.Output.W:0 (256, 1) 256\n",
            "/device:GPU:0 Discriminator2.Output.W.u_var:0 (1, 1) 1\n",
            "/device:GPU:0 Discriminator2.Output.b:0 (1,) 1\n",
            "/device:GPU:0 Discriminator.2.MHINGEGANOutput.W:0 (256, 11) 2816\n",
            "/device:GPU:0 Discriminator.2.MHINGEGANOutput.W.u_var:0 (1, 11) 11\n",
            "/device:GPU:0 Discriminator.2.MHINGEGANOutput.b:0 (11,) 11\n",
            "/device:GPU:0 Discriminator.3.resdown1.conv1.Filters:0 (3, 3, 3, 256) 6912\n",
            "/device:GPU:0 Discriminator.3.resdown1.conv1.Filters.u_var:0 (27, 1) 27\n",
            "/device:GPU:0 Discriminator.3.resdown1.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.3.resdown1.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.3.resdown1.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.3.resdown1.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.3.resdown1.conv_sc.Filters:0 (1, 1, 3, 256) 768\n",
            "/device:GPU:0 Discriminator.3.resdown1.conv_sc.Filters.u_var:0 (3, 1) 3\n",
            "/device:GPU:0 Discriminator.3.resdown1.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.3.resdown2.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.3.resdown2.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.3.resdown2.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.3.resdown2.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.3.resdown2.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.3.resdown2.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.3.resdown2.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Discriminator.3.resdown2.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Discriminator.3.resdown2.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.3.res8.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.3.res8.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.3.res8.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.3.res8.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.3.res8.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.3.res8.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.3.res16.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.3.res16.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.3.res16.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.3.res16.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.3.res16.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.3.res16.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.3.embed.W:0 (10, 256) 2560\n",
            "/device:GPU:0 Discriminator.3.embed.W.u_var:0 (10, 1) 10\n",
            "/device:GPU:0 Discriminator3.Output.W:0 (256, 1) 256\n",
            "/device:GPU:0 Discriminator3.Output.W.u_var:0 (1, 1) 1\n",
            "/device:GPU:0 Discriminator3.Output.b:0 (1,) 1\n",
            "/device:GPU:0 Discriminator.3.MHINGEGANOutput.W:0 (256, 11) 2816\n",
            "/device:GPU:0 Discriminator.3.MHINGEGANOutput.W.u_var:0 (1, 11) 11\n",
            "/device:GPU:0 Discriminator.3.MHINGEGANOutput.b:0 (11,) 11\n",
            "/device:GPU:0 Discriminator.4.resdown1.conv1.Filters:0 (3, 3, 3, 256) 6912\n",
            "/device:GPU:0 Discriminator.4.resdown1.conv1.Filters.u_var:0 (27, 1) 27\n",
            "/device:GPU:0 Discriminator.4.resdown1.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.4.resdown1.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.4.resdown1.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.4.resdown1.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.4.resdown1.conv_sc.Filters:0 (1, 1, 3, 256) 768\n",
            "/device:GPU:0 Discriminator.4.resdown1.conv_sc.Filters.u_var:0 (3, 1) 3\n",
            "/device:GPU:0 Discriminator.4.resdown1.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.4.resdown2.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.4.resdown2.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.4.resdown2.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.4.resdown2.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.4.resdown2.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.4.resdown2.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.4.resdown2.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Discriminator.4.resdown2.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Discriminator.4.resdown2.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.4.res8.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.4.res8.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.4.res8.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.4.res8.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.4.res8.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.4.res8.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.4.res16.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.4.res16.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.4.res16.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.4.res16.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.4.res16.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.4.res16.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.4.embed.W:0 (10, 256) 2560\n",
            "/device:GPU:0 Discriminator.4.embed.W.u_var:0 (10, 1) 10\n",
            "/device:GPU:0 Discriminator4.Output.W:0 (256, 1) 256\n",
            "/device:GPU:0 Discriminator4.Output.W.u_var:0 (1, 1) 1\n",
            "/device:GPU:0 Discriminator4.Output.b:0 (1,) 1\n",
            "/device:GPU:0 Discriminator.4.MHINGEGANOutput.W:0 (256, 11) 2816\n",
            "/device:GPU:0 Discriminator.4.MHINGEGANOutput.W.u_var:0 (1, 11) 11\n",
            "/device:GPU:0 Discriminator.4.MHINGEGANOutput.b:0 (11,) 11\n",
            "/device:GPU:0 Discriminator.5.resdown1.conv1.Filters:0 (3, 3, 3, 256) 6912\n",
            "/device:GPU:0 Discriminator.5.resdown1.conv1.Filters.u_var:0 (27, 1) 27\n",
            "/device:GPU:0 Discriminator.5.resdown1.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.5.resdown1.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.5.resdown1.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.5.resdown1.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.5.resdown1.conv_sc.Filters:0 (1, 1, 3, 256) 768\n",
            "/device:GPU:0 Discriminator.5.resdown1.conv_sc.Filters.u_var:0 (3, 1) 3\n",
            "/device:GPU:0 Discriminator.5.resdown1.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.5.resdown2.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.5.resdown2.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.5.resdown2.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.5.resdown2.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.5.resdown2.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.5.resdown2.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.5.resdown2.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Discriminator.5.resdown2.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Discriminator.5.resdown2.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.5.res8.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.5.res8.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.5.res8.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.5.res8.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.5.res8.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.5.res8.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.5.res16.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.5.res16.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.5.res16.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.5.res16.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.5.res16.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.5.res16.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.5.embed.W:0 (10, 256) 2560\n",
            "/device:GPU:0 Discriminator.5.embed.W.u_var:0 (10, 1) 10\n",
            "/device:GPU:0 Discriminator5.Output.W:0 (256, 1) 256\n",
            "/device:GPU:0 Discriminator5.Output.W.u_var:0 (1, 1) 1\n",
            "/device:GPU:0 Discriminator5.Output.b:0 (1,) 1\n",
            "/device:GPU:0 Discriminator.5.MHINGEGANOutput.W:0 (256, 11) 2816\n",
            "/device:GPU:0 Discriminator.5.MHINGEGANOutput.W.u_var:0 (1, 11) 11\n",
            "/device:GPU:0 Discriminator.5.MHINGEGANOutput.b:0 (11,) 11\n",
            "/device:GPU:0 Discriminator.6.resdown1.conv1.Filters:0 (3, 3, 3, 256) 6912\n",
            "/device:GPU:0 Discriminator.6.resdown1.conv1.Filters.u_var:0 (27, 1) 27\n",
            "/device:GPU:0 Discriminator.6.resdown1.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.6.resdown1.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.6.resdown1.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.6.resdown1.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.6.resdown1.conv_sc.Filters:0 (1, 1, 3, 256) 768\n",
            "/device:GPU:0 Discriminator.6.resdown1.conv_sc.Filters.u_var:0 (3, 1) 3\n",
            "/device:GPU:0 Discriminator.6.resdown1.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.6.resdown2.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.6.resdown2.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.6.resdown2.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.6.resdown2.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.6.resdown2.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.6.resdown2.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.6.resdown2.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Discriminator.6.resdown2.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Discriminator.6.resdown2.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.6.res8.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.6.res8.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.6.res8.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.6.res8.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.6.res8.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.6.res8.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.6.res16.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.6.res16.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.6.res16.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.6.res16.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.6.res16.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.6.res16.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.6.embed.W:0 (10, 256) 2560\n",
            "/device:GPU:0 Discriminator.6.embed.W.u_var:0 (10, 1) 10\n",
            "/device:GPU:0 Discriminator6.Output.W:0 (256, 1) 256\n",
            "/device:GPU:0 Discriminator6.Output.W.u_var:0 (1, 1) 1\n",
            "/device:GPU:0 Discriminator6.Output.b:0 (1,) 1\n",
            "/device:GPU:0 Discriminator.6.MHINGEGANOutput.W:0 (256, 11) 2816\n",
            "/device:GPU:0 Discriminator.6.MHINGEGANOutput.W.u_var:0 (1, 11) 11\n",
            "/device:GPU:0 Discriminator.6.MHINGEGANOutput.b:0 (11,) 11\n",
            "/device:GPU:0 Discriminator.7.resdown1.conv1.Filters:0 (3, 3, 3, 256) 6912\n",
            "/device:GPU:0 Discriminator.7.resdown1.conv1.Filters.u_var:0 (27, 1) 27\n",
            "/device:GPU:0 Discriminator.7.resdown1.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.7.resdown1.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.7.resdown1.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.7.resdown1.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.7.resdown1.conv_sc.Filters:0 (1, 1, 3, 256) 768\n",
            "/device:GPU:0 Discriminator.7.resdown1.conv_sc.Filters.u_var:0 (3, 1) 3\n",
            "/device:GPU:0 Discriminator.7.resdown1.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.7.resdown2.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.7.resdown2.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.7.resdown2.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.7.resdown2.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.7.resdown2.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.7.resdown2.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.7.resdown2.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Discriminator.7.resdown2.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Discriminator.7.resdown2.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.7.res8.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.7.res8.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.7.res8.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.7.res8.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.7.res8.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.7.res8.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.7.res16.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.7.res16.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.7.res16.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.7.res16.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.7.res16.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.7.res16.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.7.embed.W:0 (10, 256) 2560\n",
            "/device:GPU:0 Discriminator.7.embed.W.u_var:0 (10, 1) 10\n",
            "/device:GPU:0 Discriminator7.Output.W:0 (256, 1) 256\n",
            "/device:GPU:0 Discriminator7.Output.W.u_var:0 (1, 1) 1\n",
            "/device:GPU:0 Discriminator7.Output.b:0 (1,) 1\n",
            "/device:GPU:0 Discriminator.7.MHINGEGANOutput.W:0 (256, 11) 2816\n",
            "/device:GPU:0 Discriminator.7.MHINGEGANOutput.W.u_var:0 (1, 11) 11\n",
            "/device:GPU:0 Discriminator.7.MHINGEGANOutput.b:0 (11,) 11\n",
            "/device:GPU:0 Discriminator.8.resdown1.conv1.Filters:0 (3, 3, 3, 256) 6912\n",
            "/device:GPU:0 Discriminator.8.resdown1.conv1.Filters.u_var:0 (27, 1) 27\n",
            "/device:GPU:0 Discriminator.8.resdown1.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.8.resdown1.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.8.resdown1.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.8.resdown1.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.8.resdown1.conv_sc.Filters:0 (1, 1, 3, 256) 768\n",
            "/device:GPU:0 Discriminator.8.resdown1.conv_sc.Filters.u_var:0 (3, 1) 3\n",
            "/device:GPU:0 Discriminator.8.resdown1.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.8.resdown2.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.8.resdown2.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.8.resdown2.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.8.resdown2.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.8.resdown2.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.8.resdown2.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.8.resdown2.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Discriminator.8.resdown2.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Discriminator.8.resdown2.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.8.res8.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.8.res8.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.8.res8.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.8.res8.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.8.res8.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.8.res8.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.8.res16.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.8.res16.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.8.res16.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.8.res16.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.8.res16.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.8.res16.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.8.embed.W:0 (10, 256) 2560\n",
            "/device:GPU:0 Discriminator.8.embed.W.u_var:0 (10, 1) 10\n",
            "/device:GPU:0 Discriminator8.Output.W:0 (256, 1) 256\n",
            "/device:GPU:0 Discriminator8.Output.W.u_var:0 (1, 1) 1\n",
            "/device:GPU:0 Discriminator8.Output.b:0 (1,) 1\n",
            "/device:GPU:0 Discriminator.8.MHINGEGANOutput.W:0 (256, 11) 2816\n",
            "/device:GPU:0 Discriminator.8.MHINGEGANOutput.W.u_var:0 (1, 11) 11\n",
            "/device:GPU:0 Discriminator.8.MHINGEGANOutput.b:0 (11,) 11\n",
            "/device:GPU:0 Discriminator.9.resdown1.conv1.Filters:0 (3, 3, 3, 256) 6912\n",
            "/device:GPU:0 Discriminator.9.resdown1.conv1.Filters.u_var:0 (27, 1) 27\n",
            "/device:GPU:0 Discriminator.9.resdown1.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.9.resdown1.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.9.resdown1.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.9.resdown1.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.9.resdown1.conv_sc.Filters:0 (1, 1, 3, 256) 768\n",
            "/device:GPU:0 Discriminator.9.resdown1.conv_sc.Filters.u_var:0 (3, 1) 3\n",
            "/device:GPU:0 Discriminator.9.resdown1.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.9.resdown2.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.9.resdown2.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.9.resdown2.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.9.resdown2.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.9.resdown2.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.9.resdown2.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.9.resdown2.conv_sc.Filters:0 (1, 1, 256, 256) 65536\n",
            "/device:GPU:0 Discriminator.9.resdown2.conv_sc.Filters.u_var:0 (256, 1) 256\n",
            "/device:GPU:0 Discriminator.9.resdown2.conv_sc.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.9.res8.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.9.res8.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.9.res8.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.9.res8.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.9.res8.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.9.res8.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.9.res16.conv1.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.9.res16.conv1.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.9.res16.conv1.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.9.res16.conv2.Filters:0 (3, 3, 256, 256) 589824\n",
            "/device:GPU:0 Discriminator.9.res16.conv2.Filters.u_var:0 (1, 256) 256\n",
            "/device:GPU:0 Discriminator.9.res16.conv2.Biases:0 (256,) 256\n",
            "/device:GPU:0 Discriminator.9.embed.W:0 (10, 256) 2560\n",
            "/device:GPU:0 Discriminator.9.embed.W.u_var:0 (10, 1) 10\n",
            "/device:GPU:0 Discriminator9.Output.W:0 (256, 1) 256\n",
            "/device:GPU:0 Discriminator9.Output.W.u_var:0 (1, 1) 1\n",
            "/device:GPU:0 Discriminator9.Output.b:0 (1,) 1\n",
            "/device:GPU:0 Discriminator.9.MHINGEGANOutput.W:0 (256, 11) 2816\n",
            "/device:GPU:0 Discriminator.9.MHINGEGANOutput.W.u_var:0 (1, 11) 11\n",
            "/device:GPU:0 Discriminator.9.MHINGEGANOutput.b:0 (11,) 11\n",
            "G params total:85,780,830\n",
            "D params total:42,122,896\n",
            "All params total:127,903,726\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "j25FMZz7JQRl"
      },
      "source": [
        ""
      ],
      "execution_count": 28,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "WoFMiga2JQRq",
        "outputId": "75a4a80b-8969-48dd-df64-607052d69a4a",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 88
        }
      },
      "source": [
        "# Either restore the last snapshot in snapshot path if there is any or start from scratch\n",
        "g_restore_vars = [var for var in tf.global_variables() if var.name.startswith('G')]\n",
        "d_restore_vars = [var for var in tf.global_variables() if var.name.startswith('D')]\n",
        "g_saver = tf.train.Saver(g_restore_vars,keep_checkpoint_every_n_hours=2)\n",
        "d_saver = tf.train.Saver(d_restore_vars,keep_checkpoint_every_n_hours=2)\n",
        "\n",
        "def take_snapshot():\n",
        "    g_snapshot_path = os.path.join(snapshot_dir,\"iter_\"+ ('%d'%iteration).zfill(int(np.ceil(np.log10(ITERS))+1)) +\"_G.ckpt\")\n",
        "    g_saver.save(session, g_snapshot_path)\n",
        "    d_snapshot_path = os.path.join(snapshot_dir,\"iter_\"+ ('%d'%iteration).zfill(int(np.ceil(np.log10(ITERS))+1)) +\"_D.ckpt\")\n",
        "    d_saver.save(session, d_snapshot_path)\n",
        "    print(\"Snapshot created: %s\" % d_snapshot_path.replace('_D',''))\n",
        "    \n",
        "def build_model(sess):\n",
        "        #self.init_opt()\n",
        "        sess.run(tf.global_variables_initializer())\n",
        "        if g_snapshot_path.find('.ckpt') != -1:\n",
        "            g_saver.restore(sess, g_snapshot_path)\n",
        "            d_saver.restore(sess, d_snapshot_path)\n",
        "            istart = g_snapshot_path.rfind('iter_')+5\n",
        "            iend = g_snapshot_path.rfind('_')\n",
        "            _iter = g_snapshot_path[istart:iend]\n",
        "            _iter = int(_iter)\n",
        "        else:\n",
        "            print(\"Created model with fresh parameters.\")\n",
        "            _iter = 0\n",
        "        #print(\"Go to iteration %d\" % iter)\n",
        "        return _iter\n",
        "    \n",
        "# snapshot at least every 2 hours\n",
        "\n",
        "#Restore params if there are\n",
        "lib.plot._iter[0]=iteration=build_model(session)\n",
        "summary_writer = tf.summary.FileWriter(log_dir+'summaries')"
      ],
      "execution_count": 29,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "INFO:tensorflow:Restoring parameters from /content/drive/Shared drives/shared/mix-gan/logs/MIX-MHingeGAN-10G10D-CIFAR-10-conditional-CIFAR-10/snapshots/iter_055000_G.ckpt\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "INFO:tensorflow:Restoring parameters from /content/drive/Shared drives/shared/mix-gan/logs/MIX-MHingeGAN-10G10D-CIFAR-10-conditional-CIFAR-10/snapshots/iter_055000_G.ckpt\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "INFO:tensorflow:Restoring parameters from /content/drive/Shared drives/shared/mix-gan/logs/MIX-MHingeGAN-10G10D-CIFAR-10-conditional-CIFAR-10/snapshots/iter_055000_D.ckpt\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "INFO:tensorflow:Restoring parameters from /content/drive/Shared drives/shared/mix-gan/logs/MIX-MHingeGAN-10G10D-CIFAR-10-conditional-CIFAR-10/snapshots/iter_055000_D.ckpt\n"
          ],
          "name": "stderr"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "vWe55tC6JQRt"
      },
      "source": [
        "code_gradient_penalty=code_disc_cost=cl_cost=cl_real_accuracy=cl_fake_accuracy=cl_accuracy=cl_adv_accuracy=rec_loss=pred_labels=cl_train_op=tf.zeros([],dtype=dtype)"
      ],
      "execution_count": 30,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "XggdHg5SJQRw",
        "outputId": "3b09f490-9a10-40c7-ab17-60f7498f62a8",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 70
        }
      },
      "source": [
        "print(session.run(g_w))\n",
        "print(session.run(d_w))"
      ],
      "execution_count": 31,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1]\n",
            "[0.08647875 0.11372828 0.08501752 0.10806297 0.09591661 0.10738482\n",
            " 0.09358025 0.10367319 0.1122589  0.09389875]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "scrolled": true,
        "id": "eW7ID891JQRz",
        "outputId": "64fc5430-a07b-4361-f50f-a9e759a3d641",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        }
      },
      "source": [
        "train_data_batch=np.zeros([NUM_GENERATORS*BATCH_SIZE,DATA_DIM], dtype=dtype)\n",
        "train_label_batch=np.zeros(NUM_GENERATORS*BATCH_SIZE,dtype=int)\n",
        "\n",
        "# training\n",
        "\n",
        "while iteration <=ITERS:\n",
        "    running_lr=gan_learning_rate#*(1-iteration/ITERS)\n",
        "    start_time = time.time()\n",
        "    iteration+=1\n",
        "\n",
        "    # Train generator \n",
        "    for i in range(N_GENERATOR):\n",
        "        session.run(gen_train_op,\\\n",
        "        feed_dict={real_data:train_data_batch,\n",
        "        lr:running_lr, curr_iter:iteration})\n",
        "        \n",
        "    # Train critic\n",
        "    for i in range(N_CRITIC):\n",
        "        if GET_LABELS:\n",
        "            for k in range(NUM_GENERATORS):\n",
        "                train_data_batch[k*BATCH_SIZE:(k+1)*BATCH_SIZE],train_label_batch[k*BATCH_SIZE:(k+1)*BATCH_SIZE]=next(train_gen)\n",
        "        else:\n",
        "            train_data_batch=next(train_gen)\n",
        "        \n",
        "        session.run(disc_train_op,\n",
        "            feed_dict={lr: running_lr,\\\n",
        "                       real_data:train_data_batch,\\\n",
        "                       real_labels:train_label_batch, update_p:(N_CRITIC*iteration+i)%4==0 and ADA,\\\n",
        "                      }\n",
        "        )\n",
        "                    \n",
        "    if iteration<10 or iteration % log_interval == 0:\n",
        "        iter_end_time = time.time()\n",
        "        lib.plot.plot('iter_time', iter_end_time - start_time) # this is for calculating iteration time, should not include time for testing, etc.\n",
        "        warmup()\n",
        "        lib.plot.plot('warmup_time', time.time() - iter_end_time)\n",
        "        print('g_w=',session.run(g_w))\n",
        "        print('d_w=',session.run(d_w))\n",
        "        #print(session.run([d_p, indicator]))\n",
        "        generate_image(iteration)\n",
        "        _gen_cost,\\\n",
        "        _disc_real,_indicator_stat,_disc_fake, _gradient_penalty, _wass_dist, _disc_cost,\\\n",
        "        _disc_real_acc,_disc_fake_acc,_disc_acc,\\\n",
        "        _ac_cost, _ac_real_acc, _ac_fake_acc,\\\n",
        "        = session.run([ gen_cost,\\\n",
        "        disc_real, indicator_stat, disc_fake, gradient_penalty, wass_dist, disc_cost,\\\n",
        "        disc_real_accuracy,disc_fake_accuracy,disc_accuracy,\\\n",
        "        ac_cost, ac_real_acc, ac_fake_acc],\n",
        "        feed_dict={real_data:train_data_batch,real_labels:train_label_batch, }\n",
        "        )\n",
        "        \n",
        "        lib.plot.plot('gen_cost', _gen_cost)\n",
        "        # Write logs and save samples\n",
        "        if N_CRITIC>0:\n",
        "            lib.plot.plot('disc_cost', _disc_cost)\n",
        "            lib.plot.plot('gradient_penalty', _gradient_penalty)   \n",
        "            lib.plot.plot('disc_real', np.mean(_disc_real))\n",
        "            lib.plot.plot('indicator_stat', np.mean(_indicator_stat))\n",
        "            lib.plot.plot('disc_fake', np.mean(_disc_fake))\n",
        "            lib.plot.plot('disc_fake_acc', _disc_fake_acc)\n",
        "            lib.plot.plot('disc_real_acc', _disc_real_acc)\n",
        "            lib.plot.plot('disc_acc', _disc_acc)\n",
        "            lib.plot.plot('D(Xr)-D(Xf)', np.mean(_disc_real)-np.mean(_disc_fake))\n",
        "            record['D(Xr)-D(Xf)'][counter % num_counter]=np.mean(_disc_real)-np.mean(_disc_fake)\n",
        "            lib.plot.plot('Wass_dist', _wass_dist)\n",
        "            record['Wass_dist'][counter % num_counter]=_wass_dist\n",
        "            lib.plot.plot('ac_cost', _ac_cost) \n",
        "            lib.plot.plot('ac_real_acc', _ac_real_acc) \n",
        "            lib.plot.plot('ac_fake_acc', _ac_fake_acc)      \n",
        "        if iteration % (10000) == 0:  \n",
        "            if DATASET in IMAGE_DATASETS  and NUM_CHANNELS==3:\n",
        "                _inception_score = get_inception_score(BATCH_SIZE*INCEPTION_BATCHES)\n",
        "                print('Inception Score: mean=%f, std=%f'% _inception_score)\n",
        "                lib.plot.plot('inception score', _inception_score[0])\n",
        "                record['Inception score'][counter % num_counter],record['IS_std'][counter % num_counter]=_inception_score\n",
        "            for dst in ['training_set','test_set']:\n",
        "                if dst=='test_set':\n",
        "                    _gen = inf_gen('TEST') \n",
        "                else:\n",
        "                    _gen = inf_gen('TRAIN')\n",
        "                _fid = get_fid(BATCH_SIZE*FID_BATCHES,_gen)\n",
        "                record['FID_'+dst][counter % num_counter]=_fid\n",
        "                print('FID %s=%f'% (dst, _fid))\n",
        "                lib.plot.plot('fid_%s'%dst, _fid)\n",
        "            counter+=1\n",
        "\n",
        "        lib.plot.flush(log_dir)\n",
        "    lib.plot.tick()\n",
        "    # save a snapshot\n",
        "    if iteration % snapshot_interval == 0:\n",
        "        pass\n",
        "        take_snapshot()\n",
        "    "
      ],
      "execution_count": 32,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "g_w= [0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1]\n",
            "d_w= [0.0870561  0.11294837 0.08497589 0.10811818 0.09653674 0.10756413\n",
            " 0.09381814 0.10338805 0.11152417 0.09407029]\n",
            "iter 55999\titer_time=1.19, warmup_time=55.41, gen_cost=6.10, disc_cost=2.60, gradient_penalty=0.00, disc_real=-0.43, indicator_stat=2.17, disc_fake=-0.44, disc_fake_acc=0.50, disc_real_acc=0.50, disc_acc=0.50, D(Xr)-D(Xf)=0.01, Wass_dist=0.00, ac_cost=0.11, ac_real_acc=0.97, ac_fake_acc=0.07,\n",
            "g_w= [0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1]\n",
            "d_w= [0.08687949 0.11303014 0.08535716 0.1078371  0.09668914 0.10745454\n",
            " 0.09422025 0.10296305 0.11140131 0.0941678 ]\n",
            "iter 56999\titer_time=1.29, warmup_time=8.95, gen_cost=6.12, disc_cost=2.57, gradient_penalty=0.00, disc_real=-0.40, indicator_stat=2.10, disc_fake=-0.33, disc_fake_acc=0.50, disc_real_acc=0.40, disc_acc=0.45, D(Xr)-D(Xf)=-0.07, Wass_dist=0.00, ac_cost=0.06, ac_real_acc=0.99, ac_fake_acc=0.08,\n",
            "g_w= [0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1]\n",
            "d_w= [0.08688072 0.11338336 0.08544803 0.10773827 0.09662209 0.1073584\n",
            " 0.09408595 0.10356481 0.11079483 0.09412351]\n",
            "iter 57999\titer_time=1.30, warmup_time=9.24, gen_cost=6.15, disc_cost=2.55, gradient_penalty=0.00, disc_real=-0.61, indicator_stat=2.22, disc_fake=-0.47, disc_fake_acc=0.50, disc_real_acc=0.30, disc_acc=0.40, D(Xr)-D(Xf)=-0.13, Wass_dist=0.00, ac_cost=0.08, ac_real_acc=0.99, ac_fake_acc=0.05,\n",
            "g_w= [0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1]\n",
            "d_w= [0.08699365 0.11259004 0.08576047 0.10747496 0.09693633 0.1071662\n",
            " 0.09406968 0.10347254 0.11102834 0.09450782]\n",
            "iter 58999\titer_time=1.32, warmup_time=9.24, gen_cost=6.57, disc_cost=2.49, gradient_penalty=0.00, disc_real=-0.43, indicator_stat=2.46, disc_fake=-0.43, disc_fake_acc=0.50, disc_real_acc=0.40, disc_acc=0.45, D(Xr)-D(Xf)=-0.00, Wass_dist=0.00, ac_cost=0.09, ac_real_acc=0.98, ac_fake_acc=0.02,\n",
            "g_w= [0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1]\n",
            "d_w= [0.08760054 0.11260623 0.08615055 0.10716958 0.09672394 0.10715046\n",
            " 0.09455069 0.10366797 0.11056873 0.09381129]\n",
            "Calculating Inception Score with 50000 images in 10 splits\n",
            "Inception Score calculation time: 34.520468 s\n",
            "Inception Score: mean=10.125025, std=0.095543\n",
            "Calculating FID with 50000 images from each distribution\n",
            "FID calculation time: 64.616998 s\n",
            "FID training_set=4.361765\n",
            "Calculating FID with 50000 images from each distribution\n",
            "FID calculation time: 43.645229 s\n",
            "FID test_set=6.352230\n",
            "iter 59999\titer_time=1.30, warmup_time=186.45, gen_cost=6.49, disc_cost=2.54, gradient_penalty=0.00, disc_real=-0.34, indicator_stat=2.41, disc_fake=-0.33, disc_fake_acc=0.40, disc_real_acc=0.60, disc_acc=0.50, D(Xr)-D(Xf)=-0.01, Wass_dist=0.00, ac_cost=0.10, ac_real_acc=0.98, ac_fake_acc=0.04, inception score=10.13, fid_training_set=4.36, fid_test_set=6.35,\n",
            "Snapshot created: /content/drive/Shared drives/shared/mix-gan/logs/MIX-MHingeGAN-10G10D-CIFAR-10-conditional-CIFAR-10/snapshots/iter_060000.ckpt\n",
            "g_w= [0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1]\n",
            "d_w= [0.08747034 0.11211175 0.08651351 0.10687266 0.09742004 0.10726859\n",
            " 0.09441663 0.10297525 0.11062421 0.09432697]\n",
            "iter 60999\titer_time=1.31, warmup_time=9.21, gen_cost=6.42, disc_cost=2.56, gradient_penalty=0.00, disc_real=-0.57, indicator_stat=2.32, disc_fake=-0.48, disc_fake_acc=0.40, disc_real_acc=0.40, disc_acc=0.40, D(Xr)-D(Xf)=-0.09, Wass_dist=0.00, ac_cost=0.14, ac_real_acc=0.94, ac_fake_acc=0.04,\n",
            "g_w= [0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1]\n",
            "d_w= [0.08764253 0.11181451 0.08600931 0.10686383 0.09697937 0.10729016\n",
            " 0.09441341 0.1034463  0.11066797 0.09487265]\n",
            "iter 61999\titer_time=1.31, warmup_time=9.22, gen_cost=6.10, disc_cost=2.58, gradient_penalty=0.00, disc_real=-0.43, indicator_stat=2.11, disc_fake=-0.54, disc_fake_acc=0.50, disc_real_acc=0.50, disc_acc=0.50, D(Xr)-D(Xf)=0.11, Wass_dist=0.00, ac_cost=0.08, ac_real_acc=0.99, ac_fake_acc=0.06,\n",
            "g_w= [0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1]\n",
            "d_w= [0.08797996 0.11126526 0.08668305 0.10693769 0.09731329 0.10656727\n",
            " 0.09504237 0.1032889  0.11041421 0.09450805]\n",
            "iter 62999\titer_time=1.33, warmup_time=9.36, gen_cost=6.19, disc_cost=2.61, gradient_penalty=0.00, disc_real=-0.48, indicator_stat=2.27, disc_fake=-0.58, disc_fake_acc=0.60, disc_real_acc=0.50, disc_acc=0.55, D(Xr)-D(Xf)=0.10, Wass_dist=0.00, ac_cost=0.09, ac_real_acc=0.98, ac_fake_acc=0.08,\n",
            "g_w= [0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1]\n",
            "d_w= [0.08817326 0.1116889  0.08684318 0.10647271 0.09689796 0.10738868\n",
            " 0.09472214 0.10325336 0.10964122 0.09491858]\n",
            "iter 63999\titer_time=1.29, warmup_time=9.20, gen_cost=6.29, disc_cost=2.61, gradient_penalty=0.00, disc_real=-0.37, indicator_stat=2.14, disc_fake=-0.39, disc_fake_acc=0.50, disc_real_acc=0.50, disc_acc=0.50, D(Xr)-D(Xf)=0.02, Wass_dist=0.00, ac_cost=0.07, ac_real_acc=0.99, ac_fake_acc=0.09,\n",
            "g_w= [0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1]\n",
            "d_w= [0.08830804 0.11151788 0.08709838 0.10605811 0.09768573 0.10628216\n",
            " 0.0949123  0.10301875 0.11024769 0.09487103]\n",
            "iter 64999\titer_time=1.30, warmup_time=9.19, gen_cost=6.27, disc_cost=2.52, gradient_penalty=0.00, disc_real=-0.41, indicator_stat=2.21, disc_fake=-0.45, disc_fake_acc=0.50, disc_real_acc=0.60, disc_acc=0.55, D(Xr)-D(Xf)=0.04, Wass_dist=0.00, ac_cost=0.06, ac_real_acc=0.98, ac_fake_acc=0.05,\n",
            "Snapshot created: /content/drive/Shared drives/shared/mix-gan/logs/MIX-MHingeGAN-10G10D-CIFAR-10-conditional-CIFAR-10/snapshots/iter_065000.ckpt\n",
            "g_w= [0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1]\n",
            "d_w= [0.08837257 0.11127832 0.08711423 0.10674979 0.09738718 0.10638366\n",
            " 0.09516434 0.10287565 0.10950634 0.09516793]\n",
            "iter 65999\titer_time=1.28, warmup_time=8.98, gen_cost=6.51, disc_cost=2.53, gradient_penalty=0.00, disc_real=-0.42, indicator_stat=2.30, disc_fake=-0.37, disc_fake_acc=0.50, disc_real_acc=0.60, disc_acc=0.55, D(Xr)-D(Xf)=-0.05, Wass_dist=0.00, ac_cost=0.05, ac_real_acc=0.99, ac_fake_acc=0.06,\n",
            "g_w= [0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1]\n",
            "d_w= [0.08890538 0.11074651 0.08719727 0.10677045 0.09741926 0.10620663\n",
            " 0.09458643 0.10282382 0.10988948 0.09545474]\n",
            "iter 66999\titer_time=1.32, warmup_time=8.91, gen_cost=6.27, disc_cost=2.53, gradient_penalty=0.00, disc_real=-0.47, indicator_stat=2.30, disc_fake=-0.43, disc_fake_acc=0.50, disc_real_acc=0.50, disc_acc=0.50, D(Xr)-D(Xf)=-0.04, Wass_dist=0.00, ac_cost=0.06, ac_real_acc=0.99, ac_fake_acc=0.06,\n",
            "g_w= [0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1]\n",
            "d_w= [0.08859389 0.11078646 0.08805451 0.10616935 0.09790642 0.1062092\n",
            " 0.09528568 0.10286032 0.10900916 0.09512502]\n",
            "iter 67999\titer_time=1.29, warmup_time=8.94, gen_cost=6.50, disc_cost=2.53, gradient_penalty=0.00, disc_real=-0.57, indicator_stat=2.37, disc_fake=-0.37, disc_fake_acc=0.40, disc_real_acc=0.60, disc_acc=0.50, D(Xr)-D(Xf)=-0.20, Wass_dist=0.00, ac_cost=0.07, ac_real_acc=0.99, ac_fake_acc=0.06,\n",
            "g_w= [0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1]\n",
            "d_w= [0.08897422 0.11059555 0.08775914 0.10637192 0.09741405 0.10599847\n",
            " 0.09516722 0.10295848 0.10944127 0.09531967]\n",
            "iter 68999\titer_time=1.28, warmup_time=8.83, gen_cost=6.36, disc_cost=2.55, gradient_penalty=0.00, disc_real=-0.52, indicator_stat=2.24, disc_fake=-0.39, disc_fake_acc=0.50, disc_real_acc=0.50, disc_acc=0.50, D(Xr)-D(Xf)=-0.12, Wass_dist=0.00, ac_cost=0.06, ac_real_acc=0.98, ac_fake_acc=0.06,\n",
            "g_w= [0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1]\n",
            "d_w= [0.08918872 0.11032093 0.08799166 0.10592035 0.09748152 0.1066757\n",
            " 0.09538493 0.10287844 0.10906545 0.09509225]\n",
            "Calculating Inception Score with 50000 images in 10 splits\n",
            "Inception Score calculation time: 22.622643 s\n",
            "Inception Score: mean=10.080576, std=0.087943\n",
            "Calculating FID with 50000 images from each distribution\n",
            "FID calculation time: 43.113667 s\n",
            "FID training_set=4.141881\n",
            "Calculating FID with 50000 images from each distribution\n",
            "FID calculation time: 43.123504 s\n",
            "FID test_set=6.219308\n",
            "iter 69999\titer_time=1.30, warmup_time=179.85, gen_cost=6.47, disc_cost=2.52, gradient_penalty=0.00, disc_real=-0.71, indicator_stat=2.38, disc_fake=-0.44, disc_fake_acc=0.50, disc_real_acc=0.30, disc_acc=0.40, D(Xr)-D(Xf)=-0.27, Wass_dist=0.00, ac_cost=0.07, ac_real_acc=0.99, ac_fake_acc=0.05, inception score=10.08, fid_training_set=4.14, fid_test_set=6.22,\n",
            "Snapshot created: /content/drive/Shared drives/shared/mix-gan/logs/MIX-MHingeGAN-10G10D-CIFAR-10-conditional-CIFAR-10/snapshots/iter_070000.ckpt\n",
            "g_w= [0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1]\n",
            "d_w= [0.08919585 0.11070299 0.08839009 0.10585393 0.09804372 0.10620907\n",
            " 0.09525637 0.10215089 0.10928871 0.09490831]\n",
            "iter 70999\titer_time=1.29, warmup_time=9.05, gen_cost=6.39, disc_cost=2.55, gradient_penalty=0.00, disc_real=-0.48, indicator_stat=2.23, disc_fake=-0.55, disc_fake_acc=0.50, disc_real_acc=0.40, disc_acc=0.45, D(Xr)-D(Xf)=0.06, Wass_dist=0.00, ac_cost=0.06, ac_real_acc=0.99, ac_fake_acc=0.06,\n",
            "g_w= [0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1]\n",
            "d_w= [0.08901996 0.11024095 0.08842012 0.10619964 0.09761006 0.10628415\n",
            " 0.09584808 0.10226154 0.10886746 0.095248  ]\n",
            "iter 71999\titer_time=1.28, warmup_time=8.90, gen_cost=6.50, disc_cost=2.50, gradient_penalty=0.00, disc_real=-0.44, indicator_stat=2.27, disc_fake=-0.47, disc_fake_acc=0.40, disc_real_acc=0.40, disc_acc=0.40, D(Xr)-D(Xf)=0.03, Wass_dist=0.00, ac_cost=0.03, ac_real_acc=0.99, ac_fake_acc=0.05,\n",
            "g_w= [0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1]\n",
            "d_w= [0.08963054 0.10993887 0.08888214 0.10546885 0.09778501 0.10591044\n",
            " 0.0954928  0.10288408 0.10884541 0.09516189]\n",
            "iter 72999\titer_time=1.29, warmup_time=8.91, gen_cost=6.51, disc_cost=2.52, gradient_penalty=0.00, disc_real=-0.46, indicator_stat=2.42, disc_fake=-0.41, disc_fake_acc=0.50, disc_real_acc=0.50, disc_acc=0.50, D(Xr)-D(Xf)=-0.06, Wass_dist=0.00, ac_cost=0.09, ac_real_acc=0.97, ac_fake_acc=0.05,\n",
            "g_w= [0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1]\n",
            "d_w= [0.08972764 0.10968994 0.0890692  0.10503533 0.09829847 0.10552662\n",
            " 0.09577744 0.10252474 0.10893621 0.09541443]\n",
            "iter 73999\titer_time=1.28, warmup_time=9.00, gen_cost=6.57, disc_cost=2.50, gradient_penalty=0.00, disc_real=-0.54, indicator_stat=2.32, disc_fake=-0.45, disc_fake_acc=0.60, disc_real_acc=0.30, disc_acc=0.45, D(Xr)-D(Xf)=-0.09, Wass_dist=0.00, ac_cost=0.05, ac_real_acc=0.99, ac_fake_acc=0.06,\n",
            "g_w= [0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1]\n",
            "d_w= [0.09008959 0.10959245 0.08883625 0.10511547 0.09824593 0.10568362\n",
            " 0.0957507  0.10241371 0.10848521 0.09578709]\n",
            "iter 74999\titer_time=1.30, warmup_time=9.03, gen_cost=6.72, disc_cost=2.50, gradient_penalty=0.00, disc_real=-0.60, indicator_stat=2.42, disc_fake=-0.45, disc_fake_acc=0.50, disc_real_acc=0.50, disc_acc=0.50, D(Xr)-D(Xf)=-0.16, Wass_dist=0.00, ac_cost=0.03, ac_real_acc=1.00, ac_fake_acc=0.06,\n",
            "Snapshot created: /content/drive/Shared drives/shared/mix-gan/logs/MIX-MHingeGAN-10G10D-CIFAR-10-conditional-CIFAR-10/snapshots/iter_075000.ckpt\n",
            "g_w= [0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1]\n",
            "d_w= [0.09031278 0.1098004  0.08895013 0.10511987 0.09750851 0.10576724\n",
            " 0.0955911  0.10247956 0.10843715 0.09603319]\n",
            "iter 75999\titer_time=1.28, warmup_time=8.93, gen_cost=6.51, disc_cost=2.50, gradient_penalty=0.00, disc_real=-0.59, indicator_stat=2.35, disc_fake=-0.53, disc_fake_acc=0.70, disc_real_acc=0.40, disc_acc=0.55, D(Xr)-D(Xf)=-0.05, Wass_dist=0.00, ac_cost=0.04, ac_real_acc=1.00, ac_fake_acc=0.05,\n",
            "g_w= [0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1]\n",
            "d_w= [0.09012251 0.10965066 0.08888245 0.10518453 0.09785638 0.10582334\n",
            " 0.09584124 0.10269887 0.1082678  0.09567223]\n",
            "iter 76999\titer_time=1.19, warmup_time=8.82, gen_cost=6.63, disc_cost=2.50, gradient_penalty=0.00, disc_real=-0.27, indicator_stat=2.33, disc_fake=-0.22, disc_fake_acc=0.40, disc_real_acc=0.40, disc_acc=0.40, D(Xr)-D(Xf)=-0.05, Wass_dist=0.00, ac_cost=0.02, ac_real_acc=1.00, ac_fake_acc=0.07,\n",
            "g_w= [0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1]\n",
            "d_w= [0.0904785  0.10958946 0.08907504 0.10540282 0.0978175  0.10582352\n",
            " 0.09624826 0.10243278 0.10795108 0.09518096]\n",
            "iter 77999\titer_time=1.20, warmup_time=8.81, gen_cost=6.56, disc_cost=2.50, gradient_penalty=0.00, disc_real=-0.49, indicator_stat=2.34, disc_fake=-0.35, disc_fake_acc=0.50, disc_real_acc=0.40, disc_acc=0.45, D(Xr)-D(Xf)=-0.13, Wass_dist=0.00, ac_cost=0.03, ac_real_acc=1.00, ac_fake_acc=0.05,\n",
            "g_w= [0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1]\n",
            "d_w= [0.09044782 0.10958403 0.08882654 0.10501126 0.09834755 0.10529747\n",
            " 0.0960215  0.10206929 0.1084794  0.09591521]\n",
            "iter 78999\titer_time=1.20, warmup_time=9.03, gen_cost=7.00, disc_cost=2.47, gradient_penalty=0.00, disc_real=-0.43, indicator_stat=2.66, disc_fake=-0.36, disc_fake_acc=0.60, disc_real_acc=0.40, disc_acc=0.50, D(Xr)-D(Xf)=-0.07, Wass_dist=0.00, ac_cost=0.08, ac_real_acc=0.98, ac_fake_acc=0.03,\n",
            "g_w= [0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1]\n",
            "d_w= [0.09099585 0.10907577 0.08944983 0.10496537 0.09821295 0.1057335\n",
            " 0.09598742 0.10211686 0.10776984 0.09569261]\n",
            "Calculating Inception Score with 50000 images in 10 splits\n",
            "Inception Score calculation time: 20.725198 s\n",
            "Inception Score: mean=10.134542, std=0.159230\n",
            "Calculating FID with 50000 images from each distribution\n",
            "FID calculation time: 42.393343 s\n",
            "FID training_set=3.877263\n",
            "Calculating FID with 50000 images from each distribution\n",
            "FID calculation time: 42.707500 s\n",
            "FID test_set=5.988650\n",
            "iter 79999\titer_time=1.20, warmup_time=178.60, gen_cost=7.02, disc_cost=2.45, gradient_penalty=0.00, disc_real=-0.40, indicator_stat=2.71, disc_fake=-0.34, disc_fake_acc=0.50, disc_real_acc=0.50, disc_acc=0.50, D(Xr)-D(Xf)=-0.06, Wass_dist=0.00, ac_cost=0.08, ac_real_acc=0.99, ac_fake_acc=0.03, inception score=10.13, fid_training_set=3.88, fid_test_set=5.99,\n",
            "Snapshot created: /content/drive/Shared drives/shared/mix-gan/logs/MIX-MHingeGAN-10G10D-CIFAR-10-conditional-CIFAR-10/snapshots/iter_080000.ckpt\n",
            "g_w= [0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1]\n",
            "d_w= [0.09082946 0.10887073 0.08967752 0.1046555  0.09828372 0.10537666\n",
            " 0.09571024 0.10221795 0.10818831 0.09618988]\n",
            "iter 80999\titer_time=1.20, warmup_time=8.85, gen_cost=6.60, disc_cost=2.51, gradient_penalty=0.00, disc_real=-0.33, indicator_stat=2.35, disc_fake=-0.39, disc_fake_acc=0.60, disc_real_acc=0.50, disc_acc=0.55, D(Xr)-D(Xf)=0.06, Wass_dist=0.00, ac_cost=0.02, ac_real_acc=1.00, ac_fake_acc=0.07,\n",
            "g_w= [0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1]\n",
            "d_w= [0.09047128 0.10877931 0.08936368 0.10464197 0.09880812 0.10533253\n",
            " 0.09649109 0.10216766 0.10801039 0.09593391]\n",
            "iter 81999\titer_time=1.20, warmup_time=9.15, gen_cost=7.07, disc_cost=2.47, gradient_penalty=0.00, disc_real=-0.48, indicator_stat=2.52, disc_fake=-0.61, disc_fake_acc=0.60, disc_real_acc=0.30, disc_acc=0.45, D(Xr)-D(Xf)=0.13, Wass_dist=0.00, ac_cost=0.04, ac_real_acc=1.00, ac_fake_acc=0.04,\n",
            "g_w= [0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1]\n",
            "d_w= [0.09136849 0.10871971 0.08997067 0.10478279 0.09816136 0.10499033\n",
            " 0.0961845  0.10198098 0.10768861 0.09615263]\n",
            "iter 82999\titer_time=1.20, warmup_time=8.94, gen_cost=6.94, disc_cost=2.46, gradient_penalty=0.00, disc_real=-0.46, indicator_stat=2.67, disc_fake=-0.19, disc_fake_acc=0.40, disc_real_acc=0.60, disc_acc=0.50, D(Xr)-D(Xf)=-0.27, Wass_dist=0.00, ac_cost=0.07, ac_real_acc=0.98, ac_fake_acc=0.03,\n",
            "g_w= [0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1]\n",
            "d_w= [0.09122123 0.10893254 0.09029068 0.10457132 0.09800801 0.1047331\n",
            " 0.09673473 0.10193143 0.10728374 0.09629325]\n",
            "iter 83999\titer_time=1.19, warmup_time=8.92, gen_cost=6.84, disc_cost=2.52, gradient_penalty=0.00, disc_real=-0.44, indicator_stat=2.52, disc_fake=-0.34, disc_fake_acc=0.60, disc_real_acc=0.40, disc_acc=0.50, D(Xr)-D(Xf)=-0.09, Wass_dist=0.00, ac_cost=0.05, ac_real_acc=0.99, ac_fake_acc=0.07,\n",
            "g_w= [0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1]\n",
            "d_w= [0.09148154 0.10843413 0.09029934 0.10428513 0.09832447 0.10495277\n",
            " 0.09612988 0.10232978 0.10738119 0.09638182]\n",
            "iter 84999\titer_time=1.20, warmup_time=8.77, gen_cost=6.89, disc_cost=2.48, gradient_penalty=0.00, disc_real=-0.56, indicator_stat=2.55, disc_fake=-0.43, disc_fake_acc=0.60, disc_real_acc=0.30, disc_acc=0.45, D(Xr)-D(Xf)=-0.12, Wass_dist=0.00, ac_cost=0.05, ac_real_acc=0.99, ac_fake_acc=0.05,\n",
            "Snapshot created: /content/drive/Shared drives/shared/mix-gan/logs/MIX-MHingeGAN-10G10D-CIFAR-10-conditional-CIFAR-10/snapshots/iter_085000.ckpt\n",
            "g_w= [0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1]\n",
            "d_w= [0.09151396 0.10836915 0.09051929 0.10472626 0.09855936 0.1047986\n",
            " 0.09659811 0.10161631 0.10741225 0.09588667]\n",
            "iter 85999\titer_time=1.19, warmup_time=8.92, gen_cost=6.58, disc_cost=2.50, gradient_penalty=0.00, disc_real=-0.42, indicator_stat=2.43, disc_fake=-0.31, disc_fake_acc=0.50, disc_real_acc=0.40, disc_acc=0.45, D(Xr)-D(Xf)=-0.12, Wass_dist=0.00, ac_cost=0.05, ac_real_acc=0.99, ac_fake_acc=0.05,\n",
            "g_w= [0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1]\n",
            "d_w= [0.09142978 0.10859682 0.09045707 0.10424854 0.0984132  0.10462677\n",
            " 0.09679293 0.10194676 0.10741259 0.09607554]\n",
            "iter 86999\titer_time=1.20, warmup_time=8.90, gen_cost=6.93, disc_cost=2.49, gradient_penalty=0.00, disc_real=-0.76, indicator_stat=2.70, disc_fake=-0.36, disc_fake_acc=0.50, disc_real_acc=0.40, disc_acc=0.45, D(Xr)-D(Xf)=-0.40, Wass_dist=0.00, ac_cost=0.09, ac_real_acc=0.97, ac_fake_acc=0.04,\n",
            "g_w= [0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1]\n",
            "d_w= [0.09176718 0.10846843 0.09046368 0.10416082 0.09853122 0.1046971\n",
            " 0.09694138 0.10200485 0.10679347 0.09617183]\n",
            "iter 87999\titer_time=1.20, warmup_time=9.03, gen_cost=6.92, disc_cost=2.48, gradient_penalty=0.00, disc_real=-0.44, indicator_stat=2.64, disc_fake=-0.41, disc_fake_acc=0.70, disc_real_acc=0.30, disc_acc=0.50, D(Xr)-D(Xf)=-0.03, Wass_dist=0.00, ac_cost=0.07, ac_real_acc=0.98, ac_fake_acc=0.03,\n",
            "g_w= [0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1]\n",
            "d_w= [0.09175088 0.1081286  0.09110546 0.10436391 0.09868818 0.10441846\n",
            " 0.09697751 0.1015797  0.10684159 0.09614573]\n",
            "iter 88999\titer_time=1.19, warmup_time=9.00, gen_cost=6.97, disc_cost=2.50, gradient_penalty=0.00, disc_real=-0.02, indicator_stat=2.35, disc_fake=-0.06, disc_fake_acc=0.40, disc_real_acc=0.50, disc_acc=0.45, D(Xr)-D(Xf)=0.04, Wass_dist=0.00, ac_cost=0.03, ac_real_acc=1.00, ac_fake_acc=0.06,\n",
            "g_w= [0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1]\n",
            "d_w= [0.09172408 0.10807954 0.09100669 0.10419217 0.09869248 0.10460972\n",
            " 0.09723669 0.10128914 0.10690124 0.09626834]\n",
            "Calculating Inception Score with 50000 images in 10 splits\n",
            "Inception Score calculation time: 20.713572 s\n",
            "Inception Score: mean=10.259761, std=0.156534\n",
            "Calculating FID with 50000 images from each distribution\n",
            "FID calculation time: 42.710907 s\n",
            "FID training_set=3.793806\n",
            "Calculating FID with 50000 images from each distribution\n",
            "FID calculation time: 42.866868 s\n",
            "FID test_set=5.836271\n",
            "iter 89999\titer_time=1.20, warmup_time=177.71, gen_cost=6.83, disc_cost=2.50, gradient_penalty=0.00, disc_real=-0.39, indicator_stat=2.46, disc_fake=-0.49, disc_fake_acc=0.60, disc_real_acc=0.40, disc_acc=0.50, D(Xr)-D(Xf)=0.10, Wass_dist=0.00, ac_cost=0.03, ac_real_acc=0.99, ac_fake_acc=0.05, inception score=10.26, fid_training_set=3.79, fid_test_set=5.84,\n",
            "WARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/tensorflow/python/training/saver.py:971: remove_checkpoint (from tensorflow.python.training.checkpoint_management) is deprecated and will be removed in a future version.\n",
            "Instructions for updating:\n",
            "Use standard file APIs to delete files with this prefix.\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "WARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/tensorflow/python/training/saver.py:971: remove_checkpoint (from tensorflow.python.training.checkpoint_management) is deprecated and will be removed in a future version.\n",
            "Instructions for updating:\n",
            "Use standard file APIs to delete files with this prefix.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "Snapshot created: /content/drive/Shared drives/shared/mix-gan/logs/MIX-MHingeGAN-10G10D-CIFAR-10-conditional-CIFAR-10/snapshots/iter_090000.ckpt\n",
            "g_w= [0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1]\n",
            "d_w= [0.09195451 0.10775701 0.09134989 0.10413475 0.09874695 0.10450638\n",
            " 0.09673777 0.1020497  0.10661437 0.09614868]\n",
            "iter 90999\titer_time=1.21, warmup_time=8.95, gen_cost=6.85, disc_cost=2.47, gradient_penalty=0.00, disc_real=-0.57, indicator_stat=2.46, disc_fake=-0.54, disc_fake_acc=0.50, disc_real_acc=0.50, disc_acc=0.50, D(Xr)-D(Xf)=-0.03, Wass_dist=0.00, ac_cost=0.02, ac_real_acc=1.00, ac_fake_acc=0.05,\n",
            "g_w= [0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1]\n",
            "d_w= [0.09268972 0.10779069 0.09100694 0.10389937 0.09873267 0.10444316\n",
            " 0.09697393 0.10164902 0.10677837 0.09603606]\n",
            "iter 91999\titer_time=1.20, warmup_time=8.84, gen_cost=6.79, disc_cost=2.48, gradient_penalty=0.00, disc_real=-0.21, indicator_stat=2.47, disc_fake=-0.21, disc_fake_acc=0.50, disc_real_acc=0.40, disc_acc=0.45, D(Xr)-D(Xf)=-0.00, Wass_dist=0.00, ac_cost=0.03, ac_real_acc=1.00, ac_fake_acc=0.04,\n",
            "g_w= [0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1]\n",
            "d_w= [0.09248257 0.10757088 0.09115067 0.1039644  0.09885722 0.10426085\n",
            " 0.09681518 0.10181198 0.10672541 0.09636077]\n",
            "iter 92999\titer_time=1.19, warmup_time=8.86, gen_cost=7.00, disc_cost=2.41, gradient_penalty=0.00, disc_real=-0.52, indicator_stat=2.71, disc_fake=-0.35, disc_fake_acc=0.50, disc_real_acc=0.40, disc_acc=0.45, D(Xr)-D(Xf)=-0.17, Wass_dist=0.00, ac_cost=0.03, ac_real_acc=1.00, ac_fake_acc=0.02,\n",
            "g_w= [0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1]\n",
            "d_w= [0.09205355 0.10790994 0.09110072 0.10406046 0.09873014 0.10449523\n",
            " 0.09694257 0.1017543  0.10661972 0.09633334]\n",
            "iter 93999\titer_time=1.23, warmup_time=8.96, gen_cost=7.23, disc_cost=2.47, gradient_penalty=0.00, disc_real=-0.39, indicator_stat=2.64, disc_fake=-0.21, disc_fake_acc=0.60, disc_real_acc=0.40, disc_acc=0.50, D(Xr)-D(Xf)=-0.18, Wass_dist=0.00, ac_cost=0.04, ac_real_acc=0.99, ac_fake_acc=0.04,\n",
            "g_w= [0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1]\n",
            "d_w= [0.09272221 0.10744954 0.09166068 0.10388196 0.09887198 0.10393168\n",
            " 0.09710696 0.1016045  0.10636382 0.09640672]\n",
            "iter 94999\titer_time=1.20, warmup_time=9.06, gen_cost=6.63, disc_cost=2.46, gradient_penalty=0.00, disc_real=-0.36, indicator_stat=2.48, disc_fake=-0.31, disc_fake_acc=0.50, disc_real_acc=0.50, disc_acc=0.50, D(Xr)-D(Xf)=-0.06, Wass_dist=0.00, ac_cost=0.03, ac_real_acc=1.00, ac_fake_acc=0.05,\n",
            "Snapshot created: /content/drive/Shared drives/shared/mix-gan/logs/MIX-MHingeGAN-10G10D-CIFAR-10-conditional-CIFAR-10/snapshots/iter_095000.ckpt\n",
            "g_w= [0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1]\n",
            "d_w= [0.09220461 0.1073854  0.09147418 0.10398967 0.09890521 0.10438883\n",
            " 0.0968782  0.1016206  0.10677277 0.09638056]\n",
            "iter 95999\titer_time=1.21, warmup_time=9.07, gen_cost=6.73, disc_cost=2.45, gradient_penalty=0.00, disc_real=-0.30, indicator_stat=2.58, disc_fake=-0.23, disc_fake_acc=0.40, disc_real_acc=0.60, disc_acc=0.50, D(Xr)-D(Xf)=-0.07, Wass_dist=0.00, ac_cost=0.03, ac_real_acc=0.99, ac_fake_acc=0.04,\n",
            "g_w= [0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1]\n",
            "d_w= [0.0929097  0.10764789 0.09140219 0.10368786 0.09844851 0.10424937\n",
            " 0.0969569  0.10118046 0.10644015 0.09707692]\n",
            "iter 96999\titer_time=1.19, warmup_time=8.90, gen_cost=7.35, disc_cost=2.45, gradient_penalty=0.00, disc_real=-0.31, indicator_stat=2.82, disc_fake=-0.18, disc_fake_acc=0.60, disc_real_acc=0.30, disc_acc=0.45, D(Xr)-D(Xf)=-0.13, Wass_dist=0.00, ac_cost=0.03, ac_real_acc=1.00, ac_fake_acc=0.03,\n",
            "g_w= [0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1]\n",
            "d_w= [0.09275322 0.10745327 0.09167898 0.10403117 0.09872779 0.1039667\n",
            " 0.09701932 0.10144216 0.10598784 0.09693958]\n",
            "iter 97999\titer_time=1.22, warmup_time=8.94, gen_cost=7.09, disc_cost=2.43, gradient_penalty=0.00, disc_real=-0.18, indicator_stat=2.81, disc_fake=-0.17, disc_fake_acc=0.50, disc_real_acc=0.40, disc_acc=0.45, D(Xr)-D(Xf)=-0.01, Wass_dist=0.00, ac_cost=0.04, ac_real_acc=0.99, ac_fake_acc=0.03,\n",
            "g_w= [0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1]\n",
            "d_w= [0.09271691 0.10734048 0.09235499 0.10359024 0.09910996 0.10377111\n",
            " 0.09712797 0.10134126 0.10606062 0.09658653]\n",
            "iter 98999\titer_time=1.20, warmup_time=8.91, gen_cost=6.96, disc_cost=2.51, gradient_penalty=0.00, disc_real=-0.23, indicator_stat=2.55, disc_fake=-0.20, disc_fake_acc=0.50, disc_real_acc=0.50, disc_acc=0.50, D(Xr)-D(Xf)=-0.03, Wass_dist=0.00, ac_cost=0.06, ac_real_acc=0.99, ac_fake_acc=0.05,\n",
            "g_w= [0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1]\n",
            "d_w= [0.09305768 0.10743503 0.09188044 0.10366016 0.09917562 0.10367943\n",
            " 0.09720362 0.10127972 0.10604693 0.0965813 ]\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<PIL.Image.Image image mode=RGB size=320x320 at 0x7F5F6BDC7400>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "Calculating Inception Score with 50000 images in 10 splits\n",
            "Inception Score calculation time: 20.966099 s\n",
            "Inception Score: mean=10.361073, std=0.133243\n",
            "Calculating FID with 50000 images from each distribution\n",
            "FID calculation time: 43.155927 s\n",
            "FID training_set=3.575988\n",
            "Calculating FID with 50000 images from each distribution\n",
            "FID calculation time: 42.938113 s\n",
            "FID test_set=5.690977\n",
            "iter 99999\titer_time=1.22, warmup_time=182.15, gen_cost=7.10, disc_cost=2.47, gradient_penalty=0.00, disc_real=-0.32, indicator_stat=2.52, disc_fake=-0.15, disc_fake_acc=0.50, disc_real_acc=0.40, disc_acc=0.45, D(Xr)-D(Xf)=-0.17, Wass_dist=0.00, ac_cost=0.02, ac_real_acc=1.00, ac_fake_acc=0.05, inception score=10.36, fid_training_set=3.58, fid_test_set=5.69,\n",
            "Snapshot created: /content/drive/Shared drives/shared/mix-gan/logs/MIX-MHingeGAN-10G10D-CIFAR-10-conditional-CIFAR-10/snapshots/iter_100000.ckpt\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Z3XXuzPuJQR4",
        "outputId": "6c42bf6a-48f6-40e0-9ac8-4ff3cf7897f6",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        }
      },
      "source": [
        "take_snapshot()"
      ],
      "execution_count": 33,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Snapshot created: /content/drive/Shared drives/shared/mix-gan/logs/MIX-MHingeGAN-10G10D-CIFAR-10-conditional-CIFAR-10/snapshots/iter_100001.ckpt\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "VplPhnMV_QA4",
        "outputId": "a357e257-654f-435f-f2fa-fe80dae54392",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 159
        }
      },
      "source": [
        "get_inception_score(BATCH_SIZE*INCEPTION_BATCHES)\n",
        "for dst in ['training_set','test_set']:\n",
        "  _gen = inf_gen('TRAIN') if dst=='training_set' else inf_gen('TEST')\n",
        "  _fid = get_fid(BATCH_SIZE*FID_BATCHES,_gen)\n",
        "  print('FID %s=%f'% (dst, _fid))"
      ],
      "execution_count": 34,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Calculating Inception Score with 50000 images in 10 splits\n",
            "Inception Score calculation time: 20.895972 s\n",
            "Calculating FID with 50000 images from each distribution\n",
            "FID calculation time: 43.115243 s\n",
            "FID training_set=3.640233\n",
            "Calculating FID with 50000 images from each distribution\n",
            "FID calculation time: 43.286042 s\n",
            "FID test_set=5.667606\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "0EOc_a3vJQR6",
        "outputId": "bbfd640e-ac81-4064-c016-3c53877978d7",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 70
        }
      },
      "source": [
        "get_inception_score(50000)"
      ],
      "execution_count": 35,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Calculating Inception Score with 50000 images in 10 splits\n",
            "Inception Score calculation time: 20.769103 s\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(10.303837, 0.1461065)"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 35
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "8GTMNKY_JQR8",
        "outputId": "06a3ab6e-c63c-4489-bfea-71082f887e78",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 194
        }
      },
      "source": [
        "# save statistics into a csv file\n",
        "\n",
        "import csv \n",
        "import codecs\n",
        "from io import StringIO\n",
        "names.sort()\n",
        "eval_mean={}\n",
        "unbiased_std={}\n",
        "for key in record:\n",
        "    for j in range(num_counter):\n",
        "        if record[key][j]==0:\n",
        "            record[key]=record[key][:j]\n",
        "            break\n",
        "    eval_mean[key]=np.mean(record[key])\n",
        "    unbiased_std[key]=np.std(record[key],ddof=1)    \n",
        "\n",
        "class UnicodeWriter:\n",
        "    def __init__(self, f, dialect=csv.excel, encoding=\"utf-8-sig\", **kwds):\n",
        "        self.queue = StringIO()\n",
        "        self.writer = csv.writer(self.queue, dialect=dialect, **kwds)\n",
        "        self.stream = f\n",
        "        self.encoder = codecs.getincrementalencoder(encoding)()\n",
        "    def writerow(self, row):\n",
        "        '''writerow(unicode) -> None\n",
        "        This function takes a Unicode string and encodes it to the output.\n",
        "        '''\n",
        "        self.writer.writerow([s for s in row])\n",
        "        data = self.queue.getvalue()\n",
        "        data = self.encoder.encode(data)\n",
        "        self.stream.write(data)\n",
        "        self.queue.truncate(0)\n",
        "    def writerows(self, rows):\n",
        "        for row in rows:\n",
        "            self.writerow(row)\n",
        "                       \n",
        "with open(log_dir+'stats.csv', 'wb') as csvfile:\n",
        "    spamwriter = UnicodeWriter(csvfile, delimiter=',')\n",
        "    spamwriter.writerow(names)\n",
        "    line=[]\n",
        "    for i in range(len(names)):\n",
        "        line.append(u'%s ± %s'%(str(round(eval_mean[names[i]],3)),str(round(unbiased_std[names[i]],3))))\n",
        "    spamwriter.writerow(line)\n",
        "with open(log_dir+'stats.csv', 'a') as csvfile:    \n",
        "    writer=csv.writer(csvfile)\n",
        "    line=[]\n",
        "    for i in range(len(names)):\n",
        "        line.append(eval_mean[names[i]])\n",
        "    for i in range(len(names)):\n",
        "        line.append(unbiased_std[names[i]])\n",
        "    writer.writerow(line)\n"
      ],
      "execution_count": 36,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "/usr/local/lib/python3.6/dist-packages/numpy/core/_methods.py:217: RuntimeWarning: Degrees of freedom <= 0 for slice\n",
            "  keepdims=keepdims)\n",
            "/usr/local/lib/python3.6/dist-packages/numpy/core/_methods.py:209: RuntimeWarning: invalid value encountered in double_scalars\n",
            "  ret = ret.dtype.type(ret / rcount)\n",
            "/usr/local/lib/python3.6/dist-packages/numpy/core/fromnumeric.py:3335: RuntimeWarning: Mean of empty slice.\n",
            "  out=out, **kwargs)\n",
            "/usr/local/lib/python3.6/dist-packages/numpy/core/_methods.py:161: RuntimeWarning: invalid value encountered in double_scalars\n",
            "  ret = ret.dtype.type(ret / rcount)\n",
            "/usr/local/lib/python3.6/dist-packages/numpy/core/_methods.py:186: RuntimeWarning: invalid value encountered in true_divide\n",
            "  arrmean, rcount, out=arrmean, casting='unsafe', subok=False)\n"
          ],
          "name": "stderr"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "rvBM27VeJQSA"
      },
      "source": [
        "exit()"
      ],
      "execution_count": 37,
      "outputs": []
    }
  ]
}