OpenPCDet配置

OpenPCDet相关配置

亮哥提供的spconv 1.x初版，spconv 2.x版本后续更新

openpcdet依赖spconv，首先需要安装配置spconv:GitHub - traveller59/spconv: Spatial Sparse Convolution Library

---

创建anaconda虚拟环境

按照显卡驱动版本，安装conda与pytorch（建议完全通过conda安装)，此处以pytorch 1.9，cuda版本11.1为例。

conda create -n openpcdet python=3.8
conda install pytorch torchvision torchaudio cudatoolkit=11.1 -c pytorch -c nvidia

这里注意测试一下pytorch是否工作正常，使用脚本tst.py

#检查配置的cuda版本是否一致，测试不报错即可
import torch
print(torch.version.cuda)
print(torch.cuda.is_available())
print(torch.backends.cudnn.version())
a = torch.tensor([0])
print(a.cuda().device)

---

安装spconv

首先创建一个依赖项配置脚本：spconv_config.sh

#!/bin/bash

# change these path，这些一般集群share文件夹都有
# 注意spconv要求的cmake版本好像是>3.21
CUDA_ROOT=/"个人路径"/cuda-11.1
GCC=/"个人路径"/gcc/gcc-5.4
BOOST=/"个人路径"/boost_1_65_1
CMAKE=/"个人路径"/cmake3.8/bin/

export CUDA_HOME=${CUDA_ROOT}
export CC=${GCC}/bin/gcc
export CXX=${GCC}/bin/g++
export PATH=${GCC}/bin:${CUDA_ROOT}/bin:${CMAKE}:${PATH}
export CPLUS_INCLUDE_PATH=${BOOST}/include:${CPLUS_INCLUDE_PATH}
export LD_LIBRARY_PATH=${BOOST}/lib:${CUDA_ROOT}/lib64:${LD_LIBRARY_PATH}

git clone 官方spconv代码

git clone --recursive https://github.com/traveller59/spconv
# 配置环境
source spconv_config.sh
python setup.py bdist_wheel
cd dist
pip install xxx.whl

没有报错的话，开始安装openpcdet

cd OpenPCDet
python setup.py develop

至此OpenPCDet配置结束！

---

可能出现的错误

File "/user/x/libs/anaconda3/envs/test1/lib/python3.8/site-packages/torch/utils/cpp_extension.py", line 1561, in _get_cuda_arch_flags
    arch_list[-1] += '+PTX'

IndexError: list index out of range

可在编译openpcdet之前，设置环境变量

export TORCH_CUDA_ARCH_LIST="5.2 6.0 6.1 7.0 7.5 8.0 8.6+PTX"