Modified from det3d's original document.
- A rule of thumb is that your pytorch cuda version must match the cuda version of your systsem for other cuda extensions to work properly.
we have tested the following versions of OS and softwares:
- OS: Ubuntu 16.04/18.04
- Python: 3.7.9
- PyTorch: 1.9
- CUDA: 11.2
- CUDNN: 8.1.0
# basic python libraries
conda create --name polarstream python=3.7
conda activate polarstream
conda install pytorch torchvision cudatoolkit=11.2 -c pytorch
git clone https://github.com/qchenclaire/polarstream
cd polarstream
pip install -r requirements.txt
# add CenterPoint to PYTHONPATH by adding the following line to ~/.bashrc (change the path accordingly)
export PYTHONPATH="${PYTHONPATH}:PATH_TO_POLARSTREAM"conda install pytorch-scatter -c pyggit clone https://github.com/tianweiy/nuscenes-devkit
# add the following line to ~/.bashrc and reactivate bash (remember to change the PATH_TO_NUSCENES_DEVKIT value)
export PYTHONPATH="${PYTHONPATH}:PATH_TO_NUSCENES_DEVKIT/python-sdk"# set the cuda path(change the path to your own cuda location)
export PATH=/usr/local/cuda-11.2/bin:$PATH
export CUDA_PATH=/usr/local/cuda-11.2
export CUDA_HOME=/usr/local/cuda-11.2
export LD_LIBRARY_PATH=/usr/local/cuda-11.2/lib64:$LD_LIBRARY_PATH
# Note: nms build is gpu-specific. I built different nms under different environment for 1080ti and v100
bash setup.sh Note: apex build is also gpu-specific.
git clone https://github.com/NVIDIA/apex
cd apex
pip install -v --no-cache-dir --global-option="--cpp_ext" --global-option="--cuda_ext" ./pip install spconv-cu114