服务器Ubuntu16.04 CUDA10.1 cudnn7.6.3配置OpenPCdet和spconv 1.2.1,踩坑及成功配置步骤
*先创建虚拟环境、git所spconv v1.2.1(注意选择分支) 和openPCdet
conda create -n pcdet python=3.7
conda activate pcdet
git clone https://github.com/open-mmlab/OpenPCDet.git
git clone https://github.com/traveller59/spconv.git --recursive
cd spconv
git -branch <v1.2.1>
cd third_party #删除空的文件夹,重新获取源文件
rm rf cutlass
rm rf mp11
rm rf pybind11
#获取刚才删除的源文件
git clone git://github.com/NVIDIA/cutlass.git
git clone git://github.com/boostorg/mp11.git
git clone git://github.com/pybind/pybind11.git
接下来按照https://github.com/traveller59/spconv/tree/v1.2.1所写步骤安装
sudo apt-get install libboost-all-dev
python setup.py bdist_wheel
cd dist
ls
pip install spconv-xxxxxxxxxxxxx.whl#看自己的ls
如果是CUDA10.1版本确保版本为10.1.243,如果为10.1.105会在编译过程出现错误
升级更换CUDA10.1版本参考本文,和官网文件。
配置OpenPCdet就比较简单完全按照INSTALL.md就好
https://github.com/open-mmlab/OpenPCDet
注意一点
#其中kornia== 0.5.11
pip install kornia==0.5.11
结果