【SASA环境配置】SASA: Semantics-Augmented Set Abstraction for Point-based 3D Object Detection

论文题目：SASA: Semantics-Augmented Set Abstraction for Point-based 3D Object Detection
论文: https://export.arxiv.org/pdf/2201.01976.pdf
代码: https://github.com/blakechen97/SASA

完成SASA环境配置，因为代码中用到编译工具nvcc，需要安装CUDA Toolkit(nvidia)工具包，配置步骤基本和官方的差不多，步骤如下：

conda create -n sasa python=3.6
conda activate sasa

conda install pytorch==1.8.0 torchvision==0.9.0 torchaudio==0.8.0 cudatoolkit=11.1 -c pytorch -c conda-forge

(sasa) xxx@3090ti:~$ python
>>> import torch

import torch后出现报错

OSError: xxx/anaconda3/envs/sasa/lib/python3.6/site-packages/torch/lib/../../../../libcublas.so.11: symbol free_gemm_select version libcublasLt.so.11 not defined in file libcublasLt.so.11 with link time reference

最后发现是由于在此前安装cuda时在.bashrc中为其设置了LD_LIBRARY_PATH。将LD_LIBRARY_PATH中的/usr/local/cuda/lib64删除后重启即可。

>>> import torch
>>> 

https://blog.csdn.net/weixin_44510578/article/details/136649988

a. Clone this repository.

git clone https://github.com/blakechen97/SASA.git
cd SASA

b. Install spconv library.

git clone https://github.com/traveller59/spconv.git
cd spconv
git checkout v1.2.1
git submodule update --init --recursive
python setup.py bdist_wheel
pip install ./dist/spconv-1.2.1-cp36-cp36m-linux_x86_64.whl   # wheel file name may be different
cd ..

c. Install pcdet toolbox.

pip install -r requirements.txt
python setup.py develop

KITTI dataset

python -m pcdet.datasets.kitti.kitti_dataset create_kitti_infos tools/cfgs/dataset_configs/kitti_dataset.yaml

运行demo.py

Traceback (most recent call last):
  File "/home/teng/projects/SASA/tools/demo.py", line 5, in <module>
    import mayavi.mlab as mlab
ModuleNotFoundError: No module named 'mayavi'

pip install mayavi
pip install mayavi -i http://pypi.douban.com/simple --trusted-host pypi.douban.com

********************************************************************************
WARNING: Imported VTK version (9.3) does not match the one used
         to build the TVTK classes (8.1). This may cause problems.
         Please rebuild TVTK.
********************************************************************************

pip uninstall vtk==9.3
pip install vtk==8.1  -i https://pypi.tuna.tsinghua.edu.cn/simple/

ImportError: Could not import backend for traitsui.  Make sure you
        have a suitable UI toolkit like PyQt/PySide or wxPython
        installed.

pip install mayavi
pip install pyqt5

训练阶段，先进入tools，然后执行：

python train.py --cfg_file ${CONFIG_FILE}
python train.py --cfg_file cfgs/kitti_models/3dssd_sasa.yaml

如果报错"KeyError: ‘road_plane’", 将“cfgs/kitti_models/3dssd_sasa.yaml”文件中"USE_ROAD_PLANE: True"改为“USE_ROAD_PLANE: False”。

测试：

python test.py --cfg_file cfgs/kitti_models/3dssd_sasa.yaml --ckpt \
/home/xxx/projects/SASA/output/kitti_models/3dssd_sasa/default/ckpt/checkpoint_epoch_80.pth