experiments.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "import os.path\n",
    "import json\n",
    "import numpy as np\n",
    "import sys\n",
    "import importlib\n",
    "import tensorflow as tf\n",
    "import string\n",
    "import re\n",
    "from matplotlib import pyplot as plt\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "BASE_DIR = '/home/kalugin/Documents/experiments/pointnet-autoencoder'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "ROOT_DIR = BASE_DIR\n",
    "sys.path.append(BASE_DIR) # model\n",
    "sys.path.append(ROOT_DIR) # provider\n",
    "sys.path.append(os.path.join(ROOT_DIR, 'models'))\n",
    "sys.path.append(os.path.join(ROOT_DIR, 'utils'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "MODEL = importlib.import_module('model')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "DATA_PATH = os.path.join(BASE_DIR, 'data/shapenetcore_partanno_segmentation_benchmark_v0')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "sys.path.append(os.path.join('./', 'utils'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "import show3d_balls\n",
    "import tf_util"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "root = './data/shapenetcore_partanno_segmentation_benchmark_v0/'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "with open(os.path.join(root, 'train_test_split', 'shuffled_test_file_list.json'), 'r') as f:\n",
    "    test_ids = set([str(d.split('/')[2]) for d in json.load(f)])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "catfile = os.path.join(root, 'synsetoffset2category.txt')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "cat = {}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "with open(catfile, 'r') as f:\n",
    "    for line in f:\n",
    "        ls = line.strip().split()\n",
    "        cat[ls[0]] = ls[1]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "item = 'Chair'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "dir_point = os.path.join(root, cat[item], 'points')\n",
    "dir_seg = os.path.join(root, cat[item], 'points_label')\n",
    "fns = sorted(os.listdir(dir_point))\n",
    "fns = [fn for fn in fns if fn[0:-4] in test_ids]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "meta_chair = []"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "for fn in fns:\n",
    "    token = (os.path.splitext(os.path.basename(fn))[0]) \n",
    "    meta_chair.append((os.path.join(dir_point, token + '.pts'), os.path.join(dir_seg, token + '.seg')))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "704"
      ]
     },
     "execution_count": 36,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(meta_chair)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "points = [np.loadtxt(i[0]) for i in meta_chair[:1000]]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "def unit_vector(vector):\n",
    "    \"\"\" Returns the unit vector of the vector.  \"\"\"\n",
    "    return vector / np.linalg.norm(vector)\n",
    "\n",
    "def angle_between(v1, v2):\n",
    "    \"\"\" Returns the angle in radians between vectors 'v1' and 'v2'::\n",
    "\n",
    "            >>> angle_between((1, 0, 0), (0, 1, 0))\n",
    "            1.5707963267948966\n",
    "            >>> angle_between((1, 0, 0), (1, 0, 0))\n",
    "            0.0\n",
    "            >>> angle_between((1, 0, 0), (-1, 0, 0))\n",
    "            3.141592653589793\n",
    "    \"\"\"\n",
    "    v1_u = unit_vector(v1)\n",
    "    v2_u = unit_vector(v2)\n",
    "    return np.arccos(np.clip(np.dot(v1_u, v2_u), -1.0, 1.0))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "GPU_INDEX = 0"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [],
   "source": [
    "def get_model(MODEL_PATH, batch_size=1, num_point=2048):\n",
    "    with tf.Graph().as_default():\n",
    "        with tf.device('/gpu:'+str(GPU_INDEX)):\n",
    "            pointclouds_pl, labels_pl = MODEL.placeholder_inputs(batch_size, num_point)\n",
    "            is_training_pl = tf.placeholder(tf.bool, shape=())\n",
    "            pred, end_points = MODEL.get_model(pointclouds_pl, is_training_pl)\n",
    "            loss = MODEL.get_loss(pred, labels_pl, end_points)\n",
    "            \n",
    "            embedding = tf.placeholder(tf.float32, shape=(1, 1024))\n",
    "            net = tf_util.fully_connected(embedding, 1024, bn=True, is_training=False, scope='fc1', bn_decay=None)\n",
    "            net = tf_util.fully_connected(net, 1024, bn=True, is_training=False, scope='fc2', bn_decay=None)\n",
    "            net = tf_util.fully_connected(net, num_point*3, activation_fn=None, scope='fc3')\n",
    "            net = tf.reshape(net, (batch_size, num_point, 3))\n",
    "            \n",
    "            saver = tf.train.Saver()\n",
    "        # Create a session\n",
    "        print(net)\n",
    "        config = tf.ConfigProto()\n",
    "        config.gpu_options.allow_growth = True\n",
    "        config.allow_soft_placement = True\n",
    "        sess = tf.Session(config=config)\n",
    "        # Restore variables from disk.\n",
    "        saver.restore(sess, MODEL_PATH)\n",
    "        ops = {'pointclouds_pl': pointclouds_pl,\n",
    "               'labels_pl': labels_pl,\n",
    "               'is_training_pl': is_training_pl,\n",
    "               'embedding': embedding,\n",
    "               'pred': pred,\n",
    "               'loss': loss}\n",
    "        return sess, ops, net"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:From /home/kalugin/Documents/experiments/pointnet-autoencoder/models/model.py:20: The name tf.placeholder is deprecated. Please use tf.compat.v1.placeholder instead.\n",
      "\n",
      "WARNING:tensorflow:From /home/kalugin/Documents/experiments/pointnet-autoencoder/utils/tf_util.py:147: The name tf.variable_scope is deprecated. Please use tf.compat.v1.variable_scope instead.\n",
      "\n",
      "WARNING:tensorflow:From /home/kalugin/Documents/experiments/pointnet-autoencoder/utils/tf_util.py:147: The name tf.AUTO_REUSE is deprecated. Please use tf.compat.v1.AUTO_REUSE instead.\n",
      "\n",
      "WARNING:tensorflow:\n",
      "The TensorFlow contrib module will not be included in TensorFlow 2.0.\n",
      "For more information, please see:\n",
      "  * https://github.com/tensorflow/community/blob/master/rfcs/20180907-contrib-sunset.md\n",
      "  * https://github.com/tensorflow/addons\n",
      "  * https://github.com/tensorflow/io (for I/O related ops)\n",
      "If you depend on functionality not listed there, please file an issue.\n",
      "\n",
      "WARNING:tensorflow:From /home/kalugin/Documents/experiments/pointnet-autoencoder/utils/tf_util.py:21: The name tf.get_variable is deprecated. Please use tf.compat.v1.get_variable instead.\n",
      "\n",
      "WARNING:tensorflow:From /home/kalugin/Documents/experiments/pointnet-autoencoder/utils/tf_util.py:372: The name tf.nn.max_pool is deprecated. Please use tf.nn.max_pool2d instead.\n",
      "\n",
      "Tensor(\"Reshape_2:0\", shape=(1, 2048, 3), dtype=float32, device=/device:GPU:0)\n",
      "WARNING:tensorflow:From /home/kalugin/.local/lib/python2.7/site-packages/tensorflow/python/training/saver.py:1276: checkpoint_exists (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 check for files with this prefix.\n",
      "INFO:tensorflow:Restoring parameters from ./log_chair_norotation/model_4.ckpt\n"
     ]
    }
   ],
   "source": [
    "normal_sess, normal_ops, normal_decoder = get_model('./log_chair_norotation/model_4.ckpt')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Tensor(\"Reshape_2:0\", shape=(1, 2048, 3), dtype=float32, device=/device:GPU:0)\n",
      "INFO:tensorflow:Restoring parameters from ./log_chair_norotation/noised_model_2_rate_10_std_0.05.ckpt\n"
     ]
    }
   ],
   "source": [
    "noised_sess, noised_ops, noised_decoder = get_model('./log_chair_norotation/noised_model_2_rate_10_std_0.05.ckpt')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'embedding': <tf.Tensor 'Placeholder_3:0' shape=(1, 1024) dtype=float32>,\n",
       " 'is_training_pl': <tf.Tensor 'Placeholder_2:0' shape=() dtype=bool>,\n",
       " 'labels_pl': <tf.Tensor 'Placeholder_1:0' shape=(1, 2048, 3) dtype=float32>,\n",
       " 'loss': (<tf.Tensor 'mul:0' shape=() dtype=float32>,\n",
       "  {'embedding': <tf.Tensor 'Reshape:0' shape=(1, 1024) dtype=float32>,\n",
       "   'pcloss': <tf.Tensor 'Mean:0' shape=() dtype=float32>}),\n",
       " 'pointclouds_pl': <tf.Tensor 'Placeholder:0' shape=(1, 2048, 3) dtype=float32>,\n",
       " 'pred': <tf.Tensor 'Reshape_1:0' shape=(1, 2048, 3) dtype=float32>}"
      ]
     },
     "execution_count": 23,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "normal_ops"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [],
   "source": [
    "pc = points[25][:2048]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [],
   "source": [
    "from scipy.spatial.transform import Rotation as R"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [],
   "source": [
    "angles = np.arange(0, 181, 45)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([  0,  45,  90, 135, 180])"
      ]
     },
     "execution_count": 27,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "angles"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [],
   "source": [
    "rotated_pc = {}\n",
    "for i in angles:\n",
    "    for j in angles:\n",
    "        for k in angles:\n",
    "            r = R.from_euler('xyz', [i, j, j], degrees=True)\n",
    "            rotated_pc[(i, j, k)] = r.apply(pc)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[(0, 180, 135),\n",
       " (0, 0, 45),\n",
       " (90, 180, 90),\n",
       " (180, 135, 0),\n",
       " (180, 0, 90),\n",
       " (180, 90, 90),\n",
       " (45, 0, 0),\n",
       " (180, 180, 90),\n",
       " (45, 135, 180),\n",
       " (135, 135, 180),\n",
       " (0, 0, 135),\n",
       " (135, 90, 90),\n",
       " (90, 90, 135),\n",
       " (90, 135, 180),\n",
       " (180, 90, 45),\n",
       " (135, 90, 135),\n",
       " (135, 45, 135),\n",
       " (0, 90, 135),\n",
       " (135, 180, 90),\n",
       " (90, 135, 0)]"
      ]
     },
     "execution_count": 29,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "rotated_pc.keys()[:20]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [],
   "source": [
    "limits = []\n",
    "for pc in rotated_pc.values():\n",
    "    limits.append(pc.max())\n",
    "    limits.append(pc.min())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(0.4885586338436108, -0.4730755648510844)"
      ]
     },
     "execution_count": 31,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "max(limits), min(limits)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "125"
      ]
     },
     "execution_count": 32,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(rotated_pc.keys())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [],
   "source": [
    "rotated_embeddings = {}\n",
    "rotated_reconstructions = {}\n",
    "for angles, pc in rotated_pc.items():\n",
    "    embedding, reconstruction = normal_sess.run([normal_ops['loss'][1]['embedding'], normal_ops['pred']], feed_dict=\n",
    "                             {normal_ops['pointclouds_pl']: np.expand_dims(pc, axis=0),\n",
    "                              normal_ops['is_training_pl']: False})\n",
    "    rotated_embeddings[angles] = embedding\n",
    "    rotated_reconstructions[angles] = reconstruction"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(2048, 3)"
      ]
     },
     "execution_count": 34,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "rotated_reconstructions[(0, 0, 0)][0].shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 234,
   "metadata": {},
   "outputs": [
    {
     "ename": "KeyboardInterrupt",
     "evalue": "",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-234-f5d490d55a59>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0mshow3d_balls\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mshowpoints\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mrotated_pc\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mballradius\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;36m4\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
      "\u001b[0;32m/home/kalugin/Documents/experiments/pointnet-autoencoder/utils/show3d_balls.pyc\u001b[0m in \u001b[0;36mshowpoints\u001b[0;34m(xyz, c_gt, c_pred, waittime, showrot, magnifyBlue, freezerot, background, normalizecolor, ballradius)\u001b[0m\n\u001b[1;32m    104\u001b[0m         \u001b[0mcv2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mimshow\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m'show3d'\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0mshow\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    105\u001b[0m         \u001b[0;32mif\u001b[0m \u001b[0mwaittime\u001b[0m\u001b[0;34m==\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 106\u001b[0;31m             \u001b[0mcmd\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mcv2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mwaitKey\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m10\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m%\u001b[0m\u001b[0;36m256\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    107\u001b[0m         \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    108\u001b[0m             \u001b[0mcmd\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mcv2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mwaitKey\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mwaittime\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m%\u001b[0m\u001b[0;36m256\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m: "
     ]
    }
   ],
   "source": [
    "show3d_balls.showpoints(rotated_pc[(0,0,0)], ballradius=4)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## EVEN NOT NOISED MODEL CAN NOT RECONSTRUCT ROTATED OBJECT"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 243,
   "metadata": {},
   "outputs": [
    {
     "ename": "KeyboardInterrupt",
     "evalue": "",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-243-37338d6eadcb>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0mshow3d_balls\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mshowpoints\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mnp\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mrandom\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mpermutation\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mpc\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mballradius\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;36m4\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
      "\u001b[0;32m/home/kalugin/Documents/experiments/pointnet-autoencoder/utils/show3d_balls.pyc\u001b[0m in \u001b[0;36mshowpoints\u001b[0;34m(xyz, c_gt, c_pred, waittime, showrot, magnifyBlue, freezerot, background, normalizecolor, ballradius)\u001b[0m\n\u001b[1;32m    104\u001b[0m         \u001b[0mcv2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mimshow\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m'show3d'\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0mshow\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    105\u001b[0m         \u001b[0;32mif\u001b[0m \u001b[0mwaittime\u001b[0m\u001b[0;34m==\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 106\u001b[0;31m             \u001b[0mcmd\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mcv2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mwaitKey\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m10\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m%\u001b[0m\u001b[0;36m256\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    107\u001b[0m         \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    108\u001b[0m             \u001b[0mcmd\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mcv2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mwaitKey\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mwaittime\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m%\u001b[0m\u001b[0;36m256\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m: "
     ]
    }
   ],
   "source": [
    "show3d_balls.showpoints(np.random.permutation(pc), ballradius=4)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 240,
   "metadata": {},
   "outputs": [
    {
     "ename": "AttributeError",
     "evalue": "'module' object has no attribute 'permutation'",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mAttributeError\u001b[0m                            Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-240-861ceb64a437>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0mnp\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mpermutation\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mrotated_pc\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
      "\u001b[0;31mAttributeError\u001b[0m: 'module' object has no attribute 'permutation'"
     ]
    }
   ],
   "source": [
    "np.random.permutation(rotated_pc[(0,0,0)])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 248,
   "metadata": {},
   "outputs": [
    {
     "ename": "KeyboardInterrupt",
     "evalue": "",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-248-c170cd5b2917>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[1;32m      2\u001b[0m                              {normal_ops['pointclouds_pl']: np.expand_dims(rotated_pc[(0,0,0)] + 0.02, axis=0),\n\u001b[1;32m      3\u001b[0m                               normal_ops['is_training_pl']: False})\n\u001b[0;32m----> 4\u001b[0;31m \u001b[0mshow3d_balls\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mshowpoints\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mbase_reconstruction\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mballradius\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;36m4\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
      "\u001b[0;32m/home/kalugin/Documents/experiments/pointnet-autoencoder/utils/show3d_balls.pyc\u001b[0m in \u001b[0;36mshowpoints\u001b[0;34m(xyz, c_gt, c_pred, waittime, showrot, magnifyBlue, freezerot, background, normalizecolor, ballradius)\u001b[0m\n\u001b[1;32m    104\u001b[0m         \u001b[0mcv2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mimshow\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m'show3d'\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0mshow\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    105\u001b[0m         \u001b[0;32mif\u001b[0m \u001b[0mwaittime\u001b[0m\u001b[0;34m==\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 106\u001b[0;31m             \u001b[0mcmd\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mcv2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mwaitKey\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m10\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m%\u001b[0m\u001b[0;36m256\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    107\u001b[0m         \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    108\u001b[0m             \u001b[0mcmd\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mcv2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mwaitKey\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mwaittime\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m%\u001b[0m\u001b[0;36m256\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m: "
     ]
    }
   ],
   "source": [
    "base_embedding, base_reconstruction = normal_sess.run([normal_ops['loss'][1]['embedding'], normal_ops['pred']], feed_dict=\n",
    "                             {normal_ops['pointclouds_pl']: np.expand_dims(rotated_pc[(0,0,0)] + 0.02, axis=0),\n",
    "                              normal_ops['is_training_pl']: False})\n",
    "show3d_balls.showpoints(base_reconstruction[0], ballradius=4)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 237,
   "metadata": {},
   "outputs": [
    {
     "ename": "KeyboardInterrupt",
     "evalue": "",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-237-a755ee30322e>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0mshow3d_balls\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mshowpoints\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mrotated_reconstructions\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mballradius\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;36m4\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
      "\u001b[0;32m/home/kalugin/Documents/experiments/pointnet-autoencoder/utils/show3d_balls.pyc\u001b[0m in \u001b[0;36mshowpoints\u001b[0;34m(xyz, c_gt, c_pred, waittime, showrot, magnifyBlue, freezerot, background, normalizecolor, ballradius)\u001b[0m\n\u001b[1;32m    104\u001b[0m         \u001b[0mcv2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mimshow\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m'show3d'\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0mshow\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    105\u001b[0m         \u001b[0;32mif\u001b[0m \u001b[0mwaittime\u001b[0m\u001b[0;34m==\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 106\u001b[0;31m             \u001b[0mcmd\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mcv2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mwaitKey\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m10\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m%\u001b[0m\u001b[0;36m256\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    107\u001b[0m         \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    108\u001b[0m             \u001b[0mcmd\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mcv2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mwaitKey\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mwaittime\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m%\u001b[0m\u001b[0;36m256\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m: "
     ]
    }
   ],
   "source": [
    "show3d_balls.showpoints(rotated_reconstructions[(0,0,0)][0], ballradius=4)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 208,
   "metadata": {},
   "outputs": [
    {
     "ename": "KeyboardInterrupt",
     "evalue": "",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-208-df1e5e843b49>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[1;32m      2\u001b[0m                              {normal_ops['pointclouds_pl']: np.expand_dims(points[25][:2048], axis=0),\n\u001b[1;32m      3\u001b[0m                               normal_ops['is_training_pl']: False})\n\u001b[0;32m----> 4\u001b[0;31m \u001b[0mshow3d_balls\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mshowpoints\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mbase_reconstruction\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mballradius\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;36m4\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
      "\u001b[0;32m/home/kalugin/Documents/experiments/pointnet-autoencoder/utils/show3d_balls.pyc\u001b[0m in \u001b[0;36mshowpoints\u001b[0;34m(xyz, c_gt, c_pred, waittime, showrot, magnifyBlue, freezerot, background, normalizecolor, ballradius)\u001b[0m\n\u001b[1;32m    104\u001b[0m         \u001b[0mcv2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mimshow\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m'show3d'\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0mshow\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    105\u001b[0m         \u001b[0;32mif\u001b[0m \u001b[0mwaittime\u001b[0m\u001b[0;34m==\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 106\u001b[0;31m             \u001b[0mcmd\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mcv2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mwaitKey\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m10\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m%\u001b[0m\u001b[0;36m256\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    107\u001b[0m         \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    108\u001b[0m             \u001b[0mcmd\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mcv2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mwaitKey\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mwaittime\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m%\u001b[0m\u001b[0;36m256\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m: "
     ]
    }
   ],
   "source": [
    "base_embedding, base_reconstruction = normal_sess.run([normal_ops['loss'][1]['embedding'], normal_ops['pred']], feed_dict=\n",
    "                             {normal_ops['pointclouds_pl']: np.expand_dims(points[25][:2048], axis=0),\n",
    "                              normal_ops['is_training_pl']: False})\n",
    "show3d_balls.showpoints(base_reconstruction[0], ballradius=4)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## IDEAS TO ADD NOISE:\n",
    "\n",
    "## NULLIFY SOME RANDOM POINTS \n",
    "\n",
    "## ADD UNIFORME NOISE ALL OVER THE SPACE\n",
    "\n",
    "## NULLIFY SOME LITTLE PART OF THE OBJECT"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 197,
   "metadata": {},
   "outputs": [
    {
     "ename": "KeyboardInterrupt",
     "evalue": "",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-197-a3ace39b2742>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0mshow3d_balls\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mshowpoints\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mpc\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mballradius\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;36m4\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
      "\u001b[0;32m/home/kalugin/Documents/experiments/pointnet-autoencoder/utils/show3d_balls.pyc\u001b[0m in \u001b[0;36mshowpoints\u001b[0;34m(xyz, c_gt, c_pred, waittime, showrot, magnifyBlue, freezerot, background, normalizecolor, ballradius)\u001b[0m\n\u001b[1;32m    104\u001b[0m         \u001b[0mcv2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mimshow\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m'show3d'\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0mshow\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    105\u001b[0m         \u001b[0;32mif\u001b[0m \u001b[0mwaittime\u001b[0m\u001b[0;34m==\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 106\u001b[0;31m             \u001b[0mcmd\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mcv2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mwaitKey\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m10\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m%\u001b[0m\u001b[0;36m256\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    107\u001b[0m         \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    108\u001b[0m             \u001b[0mcmd\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mcv2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mwaitKey\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mwaittime\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m%\u001b[0m\u001b[0;36m256\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m: "
     ]
    }
   ],
   "source": [
    "show3d_balls.showpoints(pc, ballradius=4)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 198,
   "metadata": {},
   "outputs": [
    {
     "ename": "KeyboardInterrupt",
     "evalue": "",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-198-6efdedeea997>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0mshow3d_balls\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mshowpoints\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mrotated_reconstructions\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;36m0\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;36m0\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mballradius\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;36m4\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
      "\u001b[0;32m/home/kalugin/Documents/experiments/pointnet-autoencoder/utils/show3d_balls.pyc\u001b[0m in \u001b[0;36mshowpoints\u001b[0;34m(xyz, c_gt, c_pred, waittime, showrot, magnifyBlue, freezerot, background, normalizecolor, ballradius)\u001b[0m\n\u001b[1;32m    104\u001b[0m         \u001b[0mcv2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mimshow\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m'show3d'\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0mshow\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    105\u001b[0m         \u001b[0;32mif\u001b[0m \u001b[0mwaittime\u001b[0m\u001b[0;34m==\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 106\u001b[0;31m             \u001b[0mcmd\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mcv2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mwaitKey\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m10\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m%\u001b[0m\u001b[0;36m256\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    107\u001b[0m         \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    108\u001b[0m             \u001b[0mcmd\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mcv2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mwaitKey\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mwaittime\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m%\u001b[0m\u001b[0;36m256\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m: "
     ]
    }
   ],
   "source": [
    "show3d_balls.showpoints(rotated_reconstructions[(0, 0, 0)][0], ballradius=4)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 72,
   "metadata": {},
   "outputs": [],
   "source": [
    "rotated_embeddings = np.array([i[0] for i in rotated_embeddings)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 73,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.3000933"
      ]
     },
     "execution_count": 73,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "angle_between(rotated_embeddings[0], rotated_embeddings[1])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## COUNT NORM OF EMBEDDINGS THROUGH THE DATASET"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## MODEL WITH NOISE = 0.1 WORKS VERY BAD"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## MODEL WITH NOISE = 0.05 WORKS BAD ALSO"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 91,
   "metadata": {},
   "outputs": [],
   "source": [
    "dataset_embeddings = []\n",
    "dataset_reconstructions = []\n",
    "for pc in points:\n",
    "    embedding, reconstruction = normal_sess.run([normal_ops['loss'][1]['embedding'], normal_ops['pred']], feed_dict=\n",
    "                             {normal_ops['pointclouds_pl']: np.expand_dims(pc[:2048], axis=0),\n",
    "                              normal_ops['is_training_pl']: False})\n",
    "    dataset_embeddings.append(embedding)\n",
    "    dataset_reconstructions.append(reconstruction)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 128,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(1024,)"
      ]
     },
     "execution_count": 128,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dataset_embeddings[0][0].shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 127,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(array([ 17., 105., 165., 202., 166., 106.,  68.,  58.,  43.,  22.,  19.,\n",
       "         12.,  10.,  10.,  10.,   5.,   3.,   2.,   0.,   1.]),\n",
       " array([0.43223712, 0.7851825 , 1.1381278 , 1.4910731 , 1.8440186 ,\n",
       "        2.1969638 , 2.5499094 , 2.9028547 , 3.2558    , 3.6087453 ,\n",
       "        3.9616907 , 4.314636  , 4.6675816 , 5.020527  , 5.373472  ,\n",
       "        5.7264175 , 6.079363  , 6.432308  , 6.7852535 , 7.138199  ,\n",
       "        7.491144  ], dtype=float32),\n",
       " <a list of 20 Patch objects>)"
      ]
     },
     "execution_count": 127,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.hist(dataset_embeddings[0][0], bins=20)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 92,
   "metadata": {},
   "outputs": [],
   "source": [
    "emb_norms = np.array([np.linalg.norm(i) for i in dataset_embeddings])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 152,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(array([  1.,   1.,   0.,   0.,   2.,   3.,   3.,   2.,  10.,   9.,  19.,\n",
       "         39.,  64., 117., 117., 134., 121.,  40.,  20.,   2.]),\n",
       " array([54.404823, 55.78607 , 57.167324, 58.548576, 59.929825, 61.311073,\n",
       "        62.692326, 64.07358 , 65.45483 , 66.836075, 68.21733 , 69.59858 ,\n",
       "        70.97983 , 72.36108 , 73.742325, 75.12358 , 76.50483 , 77.88608 ,\n",
       "        79.267334, 80.64858 , 82.02983 ], dtype=float32),\n",
       " <a list of 20 Patch objects>)"
      ]
     },
     "execution_count": 152,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.hist(emb_norms, bins=20)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 249,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(704, 1, 1024)"
      ]
     },
     "execution_count": 249,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dataset_embeddings.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 250,
   "metadata": {},
   "outputs": [
    {
     "ename": "ImportError",
     "evalue": "No module named sklearn.manifold",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mImportError\u001b[0m                               Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-250-76a2661f83bf>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0;32mfrom\u001b[0m \u001b[0msklearn\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mmanifold\u001b[0m \u001b[0;32mimport\u001b[0m \u001b[0mTSNE\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
      "\u001b[0;31mImportError\u001b[0m: No module named sklearn.manifold"
     ]
    }
   ],
   "source": [
    "from sklearn.manifold import TSNE"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 251,
   "metadata": {},
   "outputs": [
    {
     "ename": "ImportError",
     "evalue": "No module named sklearn",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mImportError\u001b[0m                               Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-251-b7c74cbf5af0>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0;32mimport\u001b[0m \u001b[0msklearn\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
      "\u001b[0;31mImportError\u001b[0m: No module named sklearn"
     ]
    }
   ],
   "source": [
    "import sklearn"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 111,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(array([212, 217, 226, 230, 293, 299, 321, 345, 364, 380, 555, 592]),)"
      ]
     },
     "execution_count": 111,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.where(emb_norms < 65)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 118,
   "metadata": {},
   "outputs": [],
   "source": [
    "first, second = 270, 293"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 119,
   "metadata": {},
   "outputs": [],
   "source": [
    "middle = normal_sess.run(normal_decoder, feed_dict={normal_ops['embedding']: (dataset_embeddings[first] + dataset_embeddings[second]) / 2})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "show3d_balls.showpoints(middle[0], ballradius=4)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "two very different and specific objects (rare emb norm), middle value of emb have no sence"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 147,
   "metadata": {},
   "outputs": [],
   "source": [
    "noise = np.random.multivariate_normal(2 * np.ones(1024), 1.5 * np.eye(1024))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 148,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "76.3820544571572"
      ]
     },
     "execution_count": 148,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.linalg.norm(noise)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 164,
   "metadata": {},
   "outputs": [],
   "source": [
    "obj_from_noise = normal_sess.run(normal_decoder, feed_dict={normal_ops['embedding']: np.expand_dims(noise, axis=0)})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 150,
   "metadata": {},
   "outputs": [
    {
     "ename": "KeyboardInterrupt",
     "evalue": "",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-150-4e9c024e5e7d>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0mshow3d_balls\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mshowpoints\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mobj_from_noise\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mballradius\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;36m4\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
      "\u001b[0;32m/home/kalugin/Documents/experiments/pointnet-autoencoder/utils/show3d_balls.pyc\u001b[0m in \u001b[0;36mshowpoints\u001b[0;34m(xyz, c_gt, c_pred, waittime, showrot, magnifyBlue, freezerot, background, normalizecolor, ballradius)\u001b[0m\n\u001b[1;32m    104\u001b[0m         \u001b[0mcv2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mimshow\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m'show3d'\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0mshow\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    105\u001b[0m         \u001b[0;32mif\u001b[0m \u001b[0mwaittime\u001b[0m\u001b[0;34m==\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 106\u001b[0;31m             \u001b[0mcmd\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mcv2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mwaitKey\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m10\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m%\u001b[0m\u001b[0;36m256\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    107\u001b[0m         \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    108\u001b[0m             \u001b[0mcmd\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mcv2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mwaitKey\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mwaittime\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m%\u001b[0m\u001b[0;36m256\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m: "
     ]
    }
   ],
   "source": [
    "show3d_balls.showpoints(obj_from_noise[0], ballradius=4)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Object created from gaussian noise -- looks like just the simpliest chair"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 157,
   "metadata": {},
   "outputs": [],
   "source": [
    "dataset_embeddings = np.array(dataset_embeddings)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 162,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(1, 1024)"
      ]
     },
     "execution_count": 162,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dataset_embeddings.mean(axis=0).shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 165,
   "metadata": {},
   "outputs": [],
   "source": [
    "mean_obj = normal_sess.run(normal_decoder, feed_dict={normal_ops['embedding']: dataset_embeddings.mean(axis=0)})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "show3d_balls.showpoints(mean_obj[0], ballradius=4)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 256,
   "metadata": {},
   "outputs": [],
   "source": [
    "mean_obj[0][0] = 0"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 257,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[[ 0.        ,  0.        ,  0.        ],\n",
       "        [-0.06886346, -0.03349797, -0.00575994],\n",
       "        [ 0.28923652, -0.15293175, -0.20237601],\n",
       "        ...,\n",
       "        [-0.02978737, -0.07387929, -0.09354286],\n",
       "        [-0.13450755,  0.05299681,  0.00252825],\n",
       "        [ 0.13116962, -0.04751446, -0.2412029 ]]], dtype=float32)"
      ]
     },
     "execution_count": 257,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "mean_obj"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "mean object looks very much alike random object"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 75,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(0.0, 71.81719, 0.3000933)\n",
      "(22.5, 67.85972, 0.23926947)\n",
      "(45.0, 67.86036, 0.19990507)\n",
      "(67.5, 68.785194, 0.21464097)\n",
      "(90.0, 70.315994, 0.26075718)\n",
      "(112.5, 72.8224, 0.26063067)\n",
      "(135.0, 74.270775, 0.23970142)\n",
      "(157.5, 75.372055, 0.20148765)\n",
      "(180.0, 77.43076, 0.20187862)\n",
      "(202.5, 76.1203, 0.26414692)\n",
      "(225.0, 72.30133, 0.27314028)\n",
      "(247.5, 67.42528, 0.2748598)\n",
      "(270.0, 65.859344, 0.27856076)\n",
      "(292.5, 69.15883, 0.21207146)\n",
      "(315.0, 74.117004, 0.24882182)\n",
      "(337.5, 74.94361, 0.25918946)\n",
      "(360.0, 71.81719, None)\n"
     ]
    }
   ],
   "source": [
    "for i, v in enumerate(rotated_embeddings):\n",
    "    print(angles[i], np.linalg.norm(v), angle_between(rotated_embeddings[i], rotated_embeddings[i + 1]) if i + 1 < len(angles) else None)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 167,
   "metadata": {},
   "outputs": [
    {
     "ename": "KeyboardInterrupt",
     "evalue": "",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-167-a7802e413f8e>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0mshow3d_balls\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mshowpoints\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mpoints\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m100\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mballradius\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;36m4\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
      "\u001b[0;32m/home/kalugin/Documents/experiments/pointnet-autoencoder/utils/show3d_balls.pyc\u001b[0m in \u001b[0;36mshowpoints\u001b[0;34m(xyz, c_gt, c_pred, waittime, showrot, magnifyBlue, freezerot, background, normalizecolor, ballradius)\u001b[0m\n\u001b[1;32m    104\u001b[0m         \u001b[0mcv2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mimshow\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m'show3d'\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0mshow\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    105\u001b[0m         \u001b[0;32mif\u001b[0m \u001b[0mwaittime\u001b[0m\u001b[0;34m==\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 106\u001b[0;31m             \u001b[0mcmd\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mcv2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mwaitKey\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m10\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m%\u001b[0m\u001b[0;36m256\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    107\u001b[0m         \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    108\u001b[0m             \u001b[0mcmd\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mcv2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mwaitKey\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mwaittime\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m%\u001b[0m\u001b[0;36m256\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m: "
     ]
    }
   ],
   "source": [
    "show3d_balls.showpoints(points[100], ballradius=4)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 168,
   "metadata": {},
   "outputs": [
    {
     "ename": "KeyboardInterrupt",
     "evalue": "",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-168-697ea9fb9ad7>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0mshow3d_balls\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mshowpoints\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mpoints\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m200\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mballradius\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;36m4\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
      "\u001b[0;32m/home/kalugin/Documents/experiments/pointnet-autoencoder/utils/show3d_balls.pyc\u001b[0m in \u001b[0;36mshowpoints\u001b[0;34m(xyz, c_gt, c_pred, waittime, showrot, magnifyBlue, freezerot, background, normalizecolor, ballradius)\u001b[0m\n\u001b[1;32m    104\u001b[0m         \u001b[0mcv2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mimshow\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m'show3d'\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0mshow\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    105\u001b[0m         \u001b[0;32mif\u001b[0m \u001b[0mwaittime\u001b[0m\u001b[0;34m==\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 106\u001b[0;31m             \u001b[0mcmd\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mcv2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mwaitKey\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m10\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m%\u001b[0m\u001b[0;36m256\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    107\u001b[0m         \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    108\u001b[0m             \u001b[0mcmd\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mcv2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mwaitKey\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mwaittime\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m%\u001b[0m\u001b[0;36m256\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m: "
     ]
    }
   ],
   "source": [
    "show3d_balls.showpoints(points[200], ballradius=4)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 175,
   "metadata": {},
   "outputs": [
    {
     "ename": "KeyboardInterrupt",
     "evalue": "",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-175-e9314c54176b>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0mshow3d_balls\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mshowpoints\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mdataset_reconstructions\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m100\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mballradius\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;36m4\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
      "\u001b[0;32m/home/kalugin/Documents/experiments/pointnet-autoencoder/utils/show3d_balls.pyc\u001b[0m in \u001b[0;36mshowpoints\u001b[0;34m(xyz, c_gt, c_pred, waittime, showrot, magnifyBlue, freezerot, background, normalizecolor, ballradius)\u001b[0m\n\u001b[1;32m    104\u001b[0m         \u001b[0mcv2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mimshow\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m'show3d'\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0mshow\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    105\u001b[0m         \u001b[0;32mif\u001b[0m \u001b[0mwaittime\u001b[0m\u001b[0;34m==\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 106\u001b[0;31m             \u001b[0mcmd\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mcv2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mwaitKey\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m10\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m%\u001b[0m\u001b[0;36m256\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    107\u001b[0m         \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    108\u001b[0m             \u001b[0mcmd\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mcv2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mwaitKey\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mwaittime\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m%\u001b[0m\u001b[0;36m256\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m: "
     ]
    }
   ],
   "source": [
    "show3d_balls.showpoints(dataset_reconstructions[100][0], ballradius=4)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 176,
   "metadata": {},
   "outputs": [
    {
     "ename": "KeyboardInterrupt",
     "evalue": "",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-176-2ca9c2b3fdae>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0mshow3d_balls\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mshowpoints\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mdataset_reconstructions\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m200\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mballradius\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;36m4\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
      "\u001b[0;32m/home/kalugin/Documents/experiments/pointnet-autoencoder/utils/show3d_balls.pyc\u001b[0m in \u001b[0;36mshowpoints\u001b[0;34m(xyz, c_gt, c_pred, waittime, showrot, magnifyBlue, freezerot, background, normalizecolor, ballradius)\u001b[0m\n\u001b[1;32m    104\u001b[0m         \u001b[0mcv2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mimshow\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m'show3d'\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0mshow\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    105\u001b[0m         \u001b[0;32mif\u001b[0m \u001b[0mwaittime\u001b[0m\u001b[0;34m==\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 106\u001b[0;31m             \u001b[0mcmd\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mcv2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mwaitKey\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m10\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m%\u001b[0m\u001b[0;36m256\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    107\u001b[0m         \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    108\u001b[0m             \u001b[0mcmd\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mcv2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mwaitKey\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mwaittime\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m%\u001b[0m\u001b[0;36m256\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m: "
     ]
    }
   ],
   "source": [
    "show3d_balls.showpoints(dataset_reconstructions[200][0], ballradius=4)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 177,
   "metadata": {},
   "outputs": [],
   "source": [
    "middle_obj = normal_sess.run(normal_decoder, feed_dict={normal_ops['embedding']: (dataset_embeddings[100] + dataset_embeddings[200]) / 2})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 178,
   "metadata": {},
   "outputs": [
    {
     "ename": "KeyboardInterrupt",
     "evalue": "",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-178-d97038818635>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0mshow3d_balls\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mshowpoints\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mmiddle_obj\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mballradius\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;36m4\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
      "\u001b[0;32m/home/kalugin/Documents/experiments/pointnet-autoencoder/utils/show3d_balls.pyc\u001b[0m in \u001b[0;36mshowpoints\u001b[0;34m(xyz, c_gt, c_pred, waittime, showrot, magnifyBlue, freezerot, background, normalizecolor, ballradius)\u001b[0m\n\u001b[1;32m    104\u001b[0m         \u001b[0mcv2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mimshow\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m'show3d'\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0mshow\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    105\u001b[0m         \u001b[0;32mif\u001b[0m \u001b[0mwaittime\u001b[0m\u001b[0;34m==\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 106\u001b[0;31m             \u001b[0mcmd\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mcv2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mwaitKey\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m10\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m%\u001b[0m\u001b[0;36m256\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    107\u001b[0m         \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    108\u001b[0m             \u001b[0mcmd\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mcv2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mwaitKey\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mwaittime\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m%\u001b[0m\u001b[0;36m256\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m: "
     ]
    }
   ],
   "source": [
    "show3d_balls.showpoints(middle_obj[0], ballradius=4)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "middle object doesn't show a lot of \"middle\" propreties\n",
    "\n",
    "## BETTER TRAINED MODEL NEEDED"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "moved = np.expand_dims(np.clip(points[1][:2048, :] + 0.03, -1., 1.), axis=0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [],
   "source": [
    "embedding, reconstruction = normal_sess.run([normal_ops['loss'][1]['embedding'], normal_ops['pred']], feed_dict=\n",
    "                             {normal_ops['pointclouds_pl']: moved,\n",
    "                              normal_ops['is_training_pl']: False})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [],
   "source": [
    "pc_reconstruit = normal_sess.run(normal_decoder, feed_dict={normal_ops['embedding']: embedding})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(2048, 3)"
      ]
     },
     "execution_count": 49,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "pc_reconstruit[0].shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "show3d_balls.showpoints(points[1][:2048], ballradius=8)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "ename": "KeyboardInterrupt",
     "evalue": "",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-31-03612637ad9f>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0mshow3d_balls\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mshowpoints\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mreconstruction\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mballradius\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;36m8\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
      "\u001b[0;32m/home/kalugin/Documents/experiments/pointnet-autoencoder/utils/show3d_balls.pyc\u001b[0m in \u001b[0;36mshowpoints\u001b[0;34m(xyz, c_gt, c_pred, waittime, showrot, magnifyBlue, freezerot, background, normalizecolor, ballradius)\u001b[0m\n\u001b[1;32m    104\u001b[0m         \u001b[0mcv2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mimshow\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m'show3d'\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0mshow\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    105\u001b[0m         \u001b[0;32mif\u001b[0m \u001b[0mwaittime\u001b[0m\u001b[0;34m==\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 106\u001b[0;31m             \u001b[0mcmd\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mcv2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mwaitKey\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m10\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m%\u001b[0m\u001b[0;36m256\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    107\u001b[0m         \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    108\u001b[0m             \u001b[0mcmd\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mcv2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mwaitKey\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mwaittime\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m%\u001b[0m\u001b[0;36m256\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m: "
     ]
    }
   ],
   "source": [
    "show3d_balls.showpoints(reconstruction[0], ballradius=8)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [
    {
     "ename": "KeyboardInterrupt",
     "evalue": "",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-32-b5253976cae4>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0mshow3d_balls\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mshowpoints\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mpc_reconstruit\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mballradius\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;36m8\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
      "\u001b[0;32m/home/kalugin/Documents/experiments/pointnet-autoencoder/utils/show3d_balls.pyc\u001b[0m in \u001b[0;36mshowpoints\u001b[0;34m(xyz, c_gt, c_pred, waittime, showrot, magnifyBlue, freezerot, background, normalizecolor, ballradius)\u001b[0m\n\u001b[1;32m    104\u001b[0m         \u001b[0mcv2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mimshow\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m'show3d'\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0mshow\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    105\u001b[0m         \u001b[0;32mif\u001b[0m \u001b[0mwaittime\u001b[0m\u001b[0;34m==\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 106\u001b[0;31m             \u001b[0mcmd\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mcv2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mwaitKey\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m10\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m%\u001b[0m\u001b[0;36m256\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    107\u001b[0m         \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    108\u001b[0m             \u001b[0mcmd\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mcv2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mwaitKey\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mwaittime\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m%\u001b[0m\u001b[0;36m256\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m: "
     ]
    }
   ],
   "source": [
    "show3d_balls.showpoints(pc_reconstruit[0], ballradius=8)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [],
   "source": [
    "ph_points = np.expand_dims(points[0][:2048, :], axis=0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [],
   "source": [
    "moved = np.expand_dims(np.clip(points[1][:2048, :] + 0.03, -1., 1.), axis=0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 62,
   "metadata": {},
   "outputs": [],
   "source": [
    "feed_dict = {noised_ops['pointclouds_pl']: np.expand_dims(points[1][:2048, :], axis=0), noised_ops['is_training_pl']: False}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [],
   "source": [
    "noised_emb = normal_sess.run(normal_ops['loss'][1]['embedding'], feed_dict={normal_ops['pointclouds_pl']: np.expand_dims(points[1][:2048, :], axis=0),\n",
    "                                                              normal_ops['is_training_pl']: False})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [
    {
     "ename": "NameError",
     "evalue": "name 'noised_sess' is not defined",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mNameError\u001b[0m                                 Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-38-469c481aa781>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m noised_emb = noised_sess.run(noised_ops['loss'][1]['embedding'], feed_dict={noised_ops['pointclouds_pl']: np.expand_dims(points[1][:2048, :], axis=0), \n\u001b[0m\u001b[1;32m      2\u001b[0m                                                               \u001b[0mnoised_ops\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m'labels_pl'\u001b[0m\u001b[0;34m]\u001b[0m \u001b[0;34m:\u001b[0m\u001b[0mnp\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mexpand_dims\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mpoints\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m1\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;36m2048\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m:\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0maxis\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      3\u001b[0m                                                               noised_ops['is_training_pl']: False})\n",
      "\u001b[0;31mNameError\u001b[0m: name 'noised_sess' is not defined"
     ]
    }
   ],
   "source": [
    "noised_emb = noised_sess.run(noised_ops['loss'][1]['embedding'], feed_dict={noised_ops['pointclouds_pl']: np.expand_dims(points[1][:2048, :], axis=0), \n",
    "                                                              noised_ops['labels_pl'] :np.expand_dims(points[1][:2048, :], axis=0),\n",
    "                                                              noised_ops['is_training_pl']: False})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([-0.07849486, -0.02878492,  0.00089697])"
      ]
     },
     "execution_count": 35,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "points[0].mean(axis=0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 85,
   "metadata": {},
   "outputs": [],
   "source": [
    "emb = noised_emb[1]['embedding']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 87,
   "metadata": {},
   "outputs": [],
   "source": [
    "custom_emb = (0., {'embedding': emb, 'pcloss' : 0.}) "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 90,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<module 'model' from '/home/kalugin/Documents/experiments/pointnet-autoencoder/models/model.pyc'>"
      ]
     },
     "execution_count": 90,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "MODEL.get_model()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "\n",
    "\n",
    "output_labels = noised_sess.run(noised_ops['pred'], feed_dict={noised_ops['is_training_pl']: False, \n",
    "                                                               noised_ops['loss'] : custom_emb})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "metadata": {},
   "outputs": [
    {
     "ename": "KeyboardInterrupt",
     "evalue": "",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-59-11a0b0a5d06a>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0mshow3d_balls\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mshowpoints\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mph_points\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mballradius\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;36m8\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
      "\u001b[0;32m/home/kalugin/Documents/experiments/pointnet-autoencoder/utils/show3d_balls.pyc\u001b[0m in \u001b[0;36mshowpoints\u001b[0;34m(xyz, c_gt, c_pred, waittime, showrot, magnifyBlue, freezerot, background, normalizecolor, ballradius)\u001b[0m\n\u001b[1;32m    104\u001b[0m         \u001b[0mcv2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mimshow\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m'show3d'\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0mshow\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    105\u001b[0m         \u001b[0;32mif\u001b[0m \u001b[0mwaittime\u001b[0m\u001b[0;34m==\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 106\u001b[0;31m             \u001b[0mcmd\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mcv2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mwaitKey\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m10\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m%\u001b[0m\u001b[0;36m256\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    107\u001b[0m         \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    108\u001b[0m             \u001b[0mcmd\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mcv2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mwaitKey\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mwaittime\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m%\u001b[0m\u001b[0;36m256\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m: "
     ]
    }
   ],
   "source": [
    "show3d_balls.showpoints(ph_points[0], ballradius=8)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "metadata": {},
   "outputs": [
    {
     "ename": "KeyboardInterrupt",
     "evalue": "",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-60-bb879a2c9a34>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0mshow3d_balls\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mshowpoints\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mnoised_emb\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mballradius\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;36m8\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
      "\u001b[0;32m/home/kalugin/Documents/experiments/pointnet-autoencoder/utils/show3d_balls.pyc\u001b[0m in \u001b[0;36mshowpoints\u001b[0;34m(xyz, c_gt, c_pred, waittime, showrot, magnifyBlue, freezerot, background, normalizecolor, ballradius)\u001b[0m\n\u001b[1;32m    104\u001b[0m         \u001b[0mcv2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mimshow\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m'show3d'\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0mshow\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    105\u001b[0m         \u001b[0;32mif\u001b[0m \u001b[0mwaittime\u001b[0m\u001b[0;34m==\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 106\u001b[0;31m             \u001b[0mcmd\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mcv2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mwaitKey\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m10\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m%\u001b[0m\u001b[0;36m256\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    107\u001b[0m         \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    108\u001b[0m             \u001b[0mcmd\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mcv2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mwaitKey\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mwaittime\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m%\u001b[0m\u001b[0;36m256\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m: "
     ]
    }
   ],
   "source": [
    "show3d_balls.showpoints(noised_emb[0][0], ballradius=8)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 379,
   "metadata": {},
   "outputs": [],
   "source": [
    "batch = np.reshape(np.arange(3*2048*32), newshape=(32, 2048, 3))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 315,
   "metadata": {},
   "outputs": [],
   "source": [
    "batch[:, 0] = 0"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 380,
   "metadata": {},
   "outputs": [],
   "source": [
    "batch = batch / 150000. - 0.2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 381,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[[-0.2       , -0.19999333, -0.19998667],\n",
       "        [-0.19998   , -0.19997333, -0.19996667],\n",
       "        [-0.19996   , -0.19995333, -0.19994667],\n",
       "        ...,\n",
       "        [-0.1591    , -0.15909333, -0.15908667],\n",
       "        [-0.15908   , -0.15907333, -0.15906667],\n",
       "        [-0.15906   , -0.15905333, -0.15904667]],\n",
       "\n",
       "       [[-0.15904   , -0.15903333, -0.15902667],\n",
       "        [-0.15902   , -0.15901333, -0.15900667],\n",
       "        [-0.159     , -0.15899333, -0.15898667],\n",
       "        ...,\n",
       "        [-0.11814   , -0.11813333, -0.11812667],\n",
       "        [-0.11812   , -0.11811333, -0.11810667],\n",
       "        [-0.1181    , -0.11809333, -0.11808667]],\n",
       "\n",
       "       [[-0.11808   , -0.11807333, -0.11806667],\n",
       "        [-0.11806   , -0.11805333, -0.11804667],\n",
       "        [-0.11804   , -0.11803333, -0.11802667],\n",
       "        ...,\n",
       "        [-0.07718   , -0.07717333, -0.07716667],\n",
       "        [-0.07716   , -0.07715333, -0.07714667],\n",
       "        [-0.07714   , -0.07713333, -0.07712667]],\n",
       "\n",
       "       ...,\n",
       "\n",
       "       [[ 0.98784   ,  0.98784667,  0.98785333],\n",
       "        [ 0.98786   ,  0.98786667,  0.98787333],\n",
       "        [ 0.98788   ,  0.98788667,  0.98789333],\n",
       "        ...,\n",
       "        [ 1.02874   ,  1.02874667,  1.02875333],\n",
       "        [ 1.02876   ,  1.02876667,  1.02877333],\n",
       "        [ 1.02878   ,  1.02878667,  1.02879333]],\n",
       "\n",
       "       [[ 1.0288    ,  1.02880667,  1.02881333],\n",
       "        [ 1.02882   ,  1.02882667,  1.02883333],\n",
       "        [ 1.02884   ,  1.02884667,  1.02885333],\n",
       "        ...,\n",
       "        [ 1.0697    ,  1.06970667,  1.06971333],\n",
       "        [ 1.06972   ,  1.06972667,  1.06973333],\n",
       "        [ 1.06974   ,  1.06974667,  1.06975333]],\n",
       "\n",
       "       [[ 1.06976   ,  1.06976667,  1.06977333],\n",
       "        [ 1.06978   ,  1.06978667,  1.06979333],\n",
       "        [ 1.0698    ,  1.06980667,  1.06981333],\n",
       "        ...,\n",
       "        [ 1.11066   ,  1.11066667,  1.11067333],\n",
       "        [ 1.11068   ,  1.11068667,  1.11069333],\n",
       "        [ 1.1107    ,  1.11070667,  1.11071333]]])"
      ]
     },
     "execution_count": 381,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "batch"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 382,
   "metadata": {},
   "outputs": [],
   "source": [
    "centers = np.random.uniform(-1, 1, (32, 1, 3))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 383,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(32, 2048, 3)"
      ]
     },
     "execution_count": 383,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "batch.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 385,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(32, 1, 3)"
      ]
     },
     "execution_count": 385,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "centers.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 391,
   "metadata": {},
   "outputs": [],
   "source": [
    "c1 = np.repeat(centers, 2048, axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 392,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(32, 2048, 3)"
      ]
     },
     "execution_count": 392,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "c1.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 337,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[[-0.73360924, -0.45105486, -0.488406  ],\n",
       "        [-0.73360924, -0.45105486, -0.488406  ],\n",
       "        [-0.73360924, -0.45105486, -0.488406  ],\n",
       "        ...,\n",
       "        [-0.73360924, -0.45105486, -0.488406  ],\n",
       "        [-0.73360924, -0.45105486, -0.488406  ],\n",
       "        [-0.73360924, -0.45105486, -0.488406  ]],\n",
       "\n",
       "       [[ 0.42938142,  0.80837704,  0.62523524],\n",
       "        [ 0.42938142,  0.80837704,  0.62523524],\n",
       "        [ 0.42938142,  0.80837704,  0.62523524],\n",
       "        ...,\n",
       "        [ 0.42938142,  0.80837704,  0.62523524],\n",
       "        [ 0.42938142,  0.80837704,  0.62523524],\n",
       "        [ 0.42938142,  0.80837704,  0.62523524]],\n",
       "\n",
       "       [[-0.07960751,  0.36652193, -0.83415634],\n",
       "        [-0.07960751,  0.36652193, -0.83415634],\n",
       "        [-0.07960751,  0.36652193, -0.83415634],\n",
       "        ...,\n",
       "        [-0.07960751,  0.36652193, -0.83415634],\n",
       "        [-0.07960751,  0.36652193, -0.83415634],\n",
       "        [-0.07960751,  0.36652193, -0.83415634]],\n",
       "\n",
       "       ...,\n",
       "\n",
       "       [[ 0.77416913, -0.08697646, -0.10364255],\n",
       "        [ 0.77416913, -0.08697646, -0.10364255],\n",
       "        [ 0.77416913, -0.08697646, -0.10364255],\n",
       "        ...,\n",
       "        [ 0.77416913, -0.08697646, -0.10364255],\n",
       "        [ 0.77416913, -0.08697646, -0.10364255],\n",
       "        [ 0.77416913, -0.08697646, -0.10364255]],\n",
       "\n",
       "       [[ 0.16249448, -0.03202561,  0.48564599],\n",
       "        [ 0.16249448, -0.03202561,  0.48564599],\n",
       "        [ 0.16249448, -0.03202561,  0.48564599],\n",
       "        ...,\n",
       "        [ 0.16249448, -0.03202561,  0.48564599],\n",
       "        [ 0.16249448, -0.03202561,  0.48564599],\n",
       "        [ 0.16249448, -0.03202561,  0.48564599]],\n",
       "\n",
       "       [[ 0.49671468, -0.58863276, -0.20546211],\n",
       "        [ 0.49671468, -0.58863276, -0.20546211],\n",
       "        [ 0.49671468, -0.58863276, -0.20546211],\n",
       "        ...,\n",
       "        [ 0.49671468, -0.58863276, -0.20546211],\n",
       "        [ 0.49671468, -0.58863276, -0.20546211],\n",
       "        [ 0.49671468, -0.58863276, -0.20546211]]])"
      ]
     },
     "execution_count": 337,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "c1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 393,
   "metadata": {},
   "outputs": [],
   "source": [
    "dists = batch - c1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 394,
   "metadata": {},
   "outputs": [],
   "source": [
    "dists = np.linalg.norm(dists, axis=2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 395,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(32, 2048)"
      ]
     },
     "execution_count": 395,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dists.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 396,
   "metadata": {},
   "outputs": [],
   "source": [
    "quantiles = np.quantile(dists, 0.1, axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 397,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(32, 2048)"
      ]
     },
     "execution_count": 397,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dists.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 398,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(32,)"
      ]
     },
     "execution_count": 398,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "quantiles.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 399,
   "metadata": {},
   "outputs": [],
   "source": [
    "q_r = np.repeat(quantiles[:, np.newaxis], 2048, axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 400,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(32, 2048)"
      ]
     },
     "execution_count": 400,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "q_r.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 401,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[0.96686077, 0.96685255, 0.96684433, ..., 0.95253897, 0.9525332 ,\n",
       "        0.95252743],\n",
       "       [1.21608057, 1.21606613, 1.21605168, ..., 1.18828421, 1.18827149,\n",
       "        1.18825877],\n",
       "       [1.06635824, 1.06635072, 1.0663432 , ..., 1.05325109, 1.0532458 ,\n",
       "        1.05324052],\n",
       "       ...,\n",
       "       [1.53234921, 1.53237865, 1.53240809, ..., 1.59299004, 1.5930199 ,\n",
       "        1.59304976],\n",
       "       [1.4178889 , 1.41792247, 1.41795604, ..., 1.48664157, 1.48667524,\n",
       "        1.48670891],\n",
       "       [2.14661834, 2.14665082, 2.14668329, ..., 2.21316156, 2.21319417,\n",
       "        2.21322677]])"
      ]
     },
     "execution_count": 401,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dists"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 402,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[0.95373324, 0.95373324, 0.95373324, ..., 0.95373324, 0.95373324,\n",
       "        0.95373324],\n",
       "       [1.19088021, 1.19088021, 1.19088021, ..., 1.19088021, 1.19088021,\n",
       "        1.19088021],\n",
       "       [1.05434505, 1.05434505, 1.05434505, ..., 1.05434505, 1.05434505,\n",
       "        1.05434505],\n",
       "       ...,\n",
       "       [1.53837996, 1.53837996, 1.53837996, ..., 1.53837996, 1.53837996,\n",
       "        1.53837996],\n",
       "       [1.42476167, 1.42476167, 1.42476167, ..., 1.42476167, 1.42476167,\n",
       "        1.42476167],\n",
       "       [2.15326682, 2.15326682, 2.15326682, ..., 2.15326682, 2.15326682,\n",
       "        2.15326682]])"
      ]
     },
     "execution_count": 402,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "q_r"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 406,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(array([1435., 1230., 1230., 1230., 1435.]),\n",
       " array([ 0. ,  6.2, 12.4, 18.6, 24.8, 31. ]),\n",
       " <a list of 5 Patch objects>)"
      ]
     },
     "execution_count": 406,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.hist(np.where(dists < q_r)[0], bins=5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 410,
   "metadata": {},
   "outputs": [],
   "source": [
    "batch[np.where(dists < q_r)] = 0"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 411,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(array([205.,   0.,   0.,   0.,   0.,   0.,  71., 591., 590., 591.]),\n",
       " array([0.        , 0.0204509 , 0.0409018 , 0.0613527 , 0.08180361,\n",
       "        0.10225451, 0.12270541, 0.14315631, 0.16360721, 0.18405811,\n",
       "        0.20450901]),\n",
       " <a list of 10 Patch objects>)"
      ]
     },
     "execution_count": 411,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.hist(np.linalg.norm(batch[2],axis=1))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 419,
   "metadata": {},
   "outputs": [],
   "source": [
    "batch = np.reshape(np.arange(3*2048*32, dtype=np.float32), newshape=(32, 2048, 3))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 420,
   "metadata": {},
   "outputs": [],
   "source": [
    "batch = batch / np.max(batch) * 2 - 1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 424,
   "metadata": {},
   "outputs": [],
   "source": [
    "indices = np.random.choice(np.arange(batch.shape[1]), replace=False,\n",
    "                               size=int(batch.shape[1] * 0.1))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 429,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "204"
      ]
     },
     "execution_count": 429,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "indices.size"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 428,
   "metadata": {},
   "outputs": [],
   "source": [
    "batch[:, indices] = np.random.uniform(size=(32, 204, 3))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 2",
   "language": "python",
   "name": "python2"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.17"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}