SA-SSD复现

SA-SSD复现

环境

Ubuntu 16.04
Python 3.6
pytorch 1.1.0
CUDA 10.0
Cudnn 7.6.4

依赖

• python3.6+
• pytorch (tested on 1.1.0)
• opencv
• shapely
• mayavi
• spconv (v1.0)

安装

1. 基础安装：

 conda create --name SASSD python=3.6
 conda activate SASSD
 conda install pytorch==1.1.0 torchvision==0.3.0 cudatoolkit=10.0
 conda install shapely
 conda install opencv
 conda install scikit-image
 conda install mayavi
 conda install numba
 conda install matplotlib
 conda install Cython
 conda install terminaltables
 conda install tqdm
 pip install git+https://github.com/philferriere/cocoapi.git#subdirectory=PythonAPI
 pip install pybind11 

2. 在mmdet/ops中，编译 C++/CUDA 模块，运行如下指令：

 cd mmdet/ops/points_op
 python setup.py build_ext --inplace
 cd mmdet/ops/iou3d
 python setup.py build_ext --inplace
 cd mmdet/ops/pointnet2
 python setup.py build_ext --inplace

3. ~/.bashrc中设置环境变量:

 export NUMBAPRO_CUDA_DRIVER=/usr/lib/x86_64-linux-gnu/libcuda.so
 export NUMBAPRO_NVVM=/usr/local/cuda/nvvm/lib64/libnvvm.so
 export NUMBAPRO_LIBDEVICE=/usr/local/cuda/nvvm/libdevice
 export LD_LIBRARY_PATH=/home/sf/anaconda3/lib/python3.6/site-packages/spconv;

4. spconv安装

 sudo apt-get install libboost-all-dev
 git clone https://github.com/traveller59/spconv.git --recursive
 cd spconv && git checkout 7342772
 python setup.py bdist_wheel
 cd ./dist && pip install *

5. mmdet安装

 pip install mmcv==1.0.3
 cd mmdetection-1.2.0
 python setup.py develop

调试问题可参考：
链接: 【点云目标检测】SA-SSD 调试记录.
值得一提的是：KeyError: ‘model_state’，此问题的产生是由于使用了官方预训练模型，使用自己训练的模型并不会产生此问题，故不需修改。
最后贴一下自己训练的模型在60epoch时的的评估结果：