Installing DeepVision3D ToolBox

May 9, 2022 ยท View on GitHub

Prerequisites

  • Linux or macOS
  • Python 3.8+
  • PyTorch 1.3+
  • CUDA 10.2 / CUDA 11
  • GCC 5+

Installation

(1) Clone this repository.

git clone https://github.com/dvlab-research/DeepVision3D

(2) Build anaconda Environment.

conda create -n deepvision3d python=3.8 -y
conda activate deepvision3d

(3) Install PyTorch and torchvision following the official instructions.

E.g. 1 To install PyTorch 1.10.1 with a CUDA version of 10.2:

conda install pytorch==1.10.1 torchvision==0.11.2 torchaudio==0.10.1 cudatoolkit=10.2 -c pytorch

E.g. 2 To install PyTorch 1.10.1 with a CUDA version of 11.3:

conda install pytorch==1.10.1 torchvision==0.11.2 torchaudio==0.10.1 cudatoolkit=11.3 -c pytorch -c conda-forge

(4) Install requirements.

cd /path/to/DeepVision3D
pip install -r requirements.txt

(5) Build MMCV.

pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/{cu_version}/{torch_version}/index.html

Please specify {cu_version} and {torch_version} to your required version.

E.g. 1 To install mmcv on PyTorch 1.10.0 with CUDA 10.2:

pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/cu102/torch1.10.0/index.html

E.g. 2 To install mmcv on PyTorch 1.10.0 with CUDA 11.1:

pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/cu111/torch1.10.0/index.html

More information can be found at here

(6) Build MMDetection3D

  • Install dependencies: 
    pip install mmdet==2.22.0
pip install mmsegmentation==0.22.0
  • Build MMDetection3D. 
    cd /path/to/DeepVision3D/MMDetection3D
python setup.py develop

    NOTE: If you are using RTX3090 GPUs, please add TORCH_CUDA_ARCH_LIST=8.0+PTX like:

    cd /path/to/DeepVision3D/MMDetection3D
TORCH_CUDA_ARCH_LIST=8.0+PTX python setup.py develop

(7) Build SparseConv library and OpenPCDet (We use spconv v1.2.1).

  • Clone spconv v1.2.1:

    git clone -b v1.2.1 https://github.com/traveller59/spconv --recursive

  • Install cmake & libboost:

    conda install -c statiskit libboost-dev && conda install -c anaconda cmake

  • Build spconv library:

    cd spconv && python setup.py bdist_wheel

    NOTE: If you are using RTX3090 GPUs, please add TORCH_CUDA_ARCH_LIST=8.0+PTX:

    cd spconv && TORCH_CUDA_ARCH_LIST=8.0+PTX python setup.py bdist_wheel

  • Install spconv:

    cd ./dist && pip install spconv-1.2.1-cp37-cp37m-linux_x86_64.whl

  • Install OpenPCDet:

    cd /path/to/DeepVision3D/OpenPCDet && python setup.py develop

    NOTE: If you are using RTX3090 GPUs, please add TORCH_CUDA_ARCH_LIST=8.0+PTX:

    cd /path/to/DeepVision3D/OpenPCDet && TORCH_CUDA_ARCH_LIST=8.0+PTX python setup.py develop

(8) Build EQNet.

cd /path/to/DeepVision3D/EQNet && python setup.py develop

NOTE: If you are using RTX3090 GPUs, please add TORCH_CUDA_ARCH_LIST=8.0+PTX:

cd /path/to/DeepVision3D/EQNet && TORCH_CUDA_ARCH_LIST=8.0+PTX python setup.py develop

(9) Build DVSegmentation.

  • Install Google Hashmap: 
    conda install -c bioconda google-sparsehash
  • Build DVSegmentation (Please specify the include_dirs in setup.py to your anaconda include path i.e. /path/to/anaconda3/envs/deepvision3d/include/): 
    cd /path/to/DeepVision3D/DVSegmentation/ops && python setup.py install

    NOTE: If you are using RTX3090 GPUs, please add TORCH_CUDA_ARCH_LIST=8.0+PTX:

    cd /path/to/DeepVision3D/DVSegmentation/ops && TORCH_CUDA_ARCH_LIST=8.0+PTX python setup.py install