Train

python config/conf_pcd.py

Update

2021.8.29: 添加了各种test 函数，修改了数据加载类，修改了深度图像网络，添加了稀疏网络处理模块,模块测试通过。
2021.9.01: 更新 了训练流程，配置，日志模块，ponit++ backbone 待修改，spbackbone 可以用。

Usage

python train.py $config_file_path$

Question and Description :

Q1: 读取点云没有使用降采样，open3d 降采样发现每次得到的点数维度都不一样。既然使用了稀疏处理，就 不用降采样了。，这样就可以解决batch_size 只能为1 的问题，之前好像有一个batch_size 大于1 无法训练，不知道是不是这个原因？（2021.8.29）

A1：

Q2: 维度关系，不确定现在的维度合并 合不合理？ 深度图像ResNet34 网络采用得到 [Batch, 512, 1, 1], 稀疏网络 得到[10, 15232] 。 合并得到的向量越长，self_attention 是不是处理的内存和参数是不是越来越大？是否足够影响到需要我们在乎这个问题？（2021.8.29）

A2：

D1 : 关于稀疏模块的处理，点云读取得到[batch_size, 307200, 3] , 构建dense data 为 [batch_size, 512,600,3]（2021.8.29） D2 : 目前的网络丢弃了 FC层，直接进行合并。（2021.8.29）

2. Install

install cndnn

conda install -c conda-forge cudnn

export CUDNN_LIBRARY_PATH="/home/fangxu/anaconda3/envs/torch/lib": $CUDNN_LIBRARY_PATH
export CUDNN_INCLUDE_PATH="/home/fangxu/anaconda3/envs/torch/include":$CUDNN_INCLUDE_PATH

modify cache file

traveller59/spconv#277

And fixed it by pointing CMake to the appropriate directories in CMakeCache.txt (for me it's in build/temp.linux-x86_64-3.6/CMakeCache.txt). To do this look for the following lines in your CMakeCache.txt and change them to your cuda / cudnn paths, for example:

//Folder containing NVIDIA cuDNN header files
CUDNN_INCLUDE_DIR:FILEPATH=/home/c2/anaconda3/envs/cp2/include

//Path to a file.
CUDNN_INCLUDE_PATH:PATH=/home/c2/anaconda3/envs/cp2/include

//Path to the cudnn library file (e.g., libcudnn.so)
CUDNN_LIBRARY:FILEPATH=/home/c2/anaconda3/envs/cp2/lib/libcudnn.so

//Path to a library.
CUDNN_LIBRARY_PATH:FILEPATH=/home/c2/anaconda3/envs/cp2/lib/libcudnn.so

3. Note

强制合并代码 git pull --rebase origin main

Resource : https://github.com/youngguncho/HourglassPose-Pytorch

https://github.com/V-Soboleva/PoseNet

https://github.com/jac99/MinkLoc3D

https://github.com/jac99/MinkLoc3Dv2

学习率衰减策略：https://blog.csdn.net/a906958671/article/details/104967447

cmake -DCMAKE_BUILD_TYPE=Release -D CMAKE_INSTALL_PREFIX=/usr/local/opencv34 -D WITH_CUDA=OFF -D BUILD_TIFF=ON -D WITH_GTK=ON -D WITH_QT=OFF -D PYTHON3_EXECUTABLE=/home/fangxu/anaconda3/envs/det/bin/python
-D BUILD_opencv_hdf=OFF
-D OPENCV_EXTRA_MODULES_PATH=../opencv_contrib/modules ..

/home/fangxu/anaconda3/envs/det/bin/python

-D PYTHON3_INCLUDE_DIR=/home/fangxu/anaconda3/envs/det/include
-D PYTHON3_INCLUDE_DIR=/home/fangxu/anaconda3/envs/det/include
-D PYTHON_LIBRARY=/home/fangxu/anaconda3/envs/det/lib/libpython3.8.so
-D PYTHON3_NUMPY_INCLUDE_DIRS=/home/fangxu/anaconda3/envs/det/lib/python3.8/site-packages/numpy/core/include \

-D PYTHON_LIBRARY=/home/fangxu/anaconda3/envs/det/lib/libpython3.8.so
-D PYTHON3_NUMPY_INCLUDE_DIRS=/home/fangxu/anaconda3/envs/det/lib/python3.8/site-packages/numpy/core/include
-D PYTHON3_INCLUDE_DIR=/home/fangxu/anaconda3/envs/det/include
-D PYTHON_INCLUDE_DIR2=/home/fangxu/anaconda3/envs/det/include/python3.8 \

-D PYTHON3_INCLUDE_DIR=/home/fangxu/anaconda3/envs/det/include/python3.8
-D PYTHON3_NUMPY_INCLUDE_DIRS=/home/fangxu/anaconda3/envs/det/lib/python3.8/site-packages/numpy/core/include \

cmake -DCMAKE_BUILD_TYPE=Release -D CMAKE_INSTALL_PREFIX=/usr/local/opencv34 -D WITH_CUDA=OFF -D BUILD_TIFF=ON -D WITH_GTK=ON -D WITH_QT=OFF -D PYTHON3_EXECUTABLE=/home/fangxu/anaconda3/bin/python
-D BUILD_opencv_hdf=OFF
-D OPENCV_EXTRA_MODULES_PATH=../opencv_contrib/modules ..