Some tips about environment configuration

[DY-ATL]

Before pip install -r requirements.txt, I need to

1. Add yacs to requirements.txt
2. Specify the version of torchsparse git+https://github.com/mit-han-lab/[email protected] in requirements.txt or you cannot import PointTensor in gru_fusion.py.
3. sudo apt-get install libsparsehash-dev libeigen3-dev or some file cannot be included.
4. Add CUDA path to the environment variable (add this line to ~/.bashrc and source ~/.bashrc to make it effective immediately)

export LIBRARY_PATH="/usr/local/cuda/lib64/stubs:$LIBRARY_PATH"

[DY-ATL]

It's recommended that use pip install . instead of python setup.py install in third_party/install.sh.

[xiaoc57]

hello! these tips are very useful. This issue should be referenced in the pull request.

[westlake-moonlight]

补充一条，在安装 requirements.txt 之前，要使用 pip install numpy==1.26.0 ，安装 1.26 版本的 Numpy 。
否则后续 Building wheel for torchsparse (setup.py) 阶段可能会出错，因为默认会安装的 Numpy 2.0 以上版本，导致冲突。