Reference：
PVRCNN环境搭建
环境框架：OpenPCDet
点击主页Installation
References:
instructions on how to successfully build spconv 1.0 #152
设置spconv的CMakelists
spconv1.2.1参考 README
conda虚拟环境创建cuda版本要与本机一致（如工控机为10.0）
cuda10.1+cudnn7.6.5+spconv1.2.1+pytorch1.3
conda create --name PVRCNN_NEW python=3.7 pytorch=1.3 cudatoolkit=10.1 cudatoolkit-dev=10.1 cmake --channel pytorch --channel=conda-forge
conda activate PVRCNN_NEW conda install cudnn conda install boost
Error:
PackagesNotFoundError: The following packages are not available from current channels（换成国内源）
下载spconv1.2.1（git clone）spconv1.2.1的位置在此处
下载zip
修改CMakeLists安装spconv时由于基本的路径都是在home自己的用户名下，所以需要更改。参考博客：3D目标检测算法配置–CIA-SSD与spconv1.2的安装注意spconv1.2.1的位置在此处
CUDA_ROOT=/home/qinjia/Software/anaconda3/envs/PVRCNN_NEW python setup.py bdist_wheel cd dist/ pip install *
Test spconv via:
python -c 'import spconv'
(should just return and not raise any errors)
1 pybind11找不到
pybind11地址
进入third_party,删除pybind11空目录，重新git clone即可。
若spconv安装出错，可尝试复制成功的目录，直接进入dist安装。
PCDetReference：https://blog.csdn.net/weixin_44579633/article/details/107820311
安装依赖进入到OpenPCDet目录下，安装所有依赖
cd OpenPCDet pip install -r requirements.txt
安装
运行setup.py来安装pcdet
python setup.py develop
检验
检验是否安装成功，无报错即可
python import pcdet
Test：
python -c ‘import pcdet’
注意
pcdet的环境可能会与spconv环境冲突，如pytorch版本不一致，导致两个其中一个不能运行，以下是推荐的requirements文件：
numpy
torch==1.3.1
numba
tensorboardX
easydict
pyyaml
scikit-image
tqdm
kornia
torchvision
如图，我直接执行pcdet的setup，就会报版本错误，numpy过高
执行requirements，发现pytorch被改为1.3.0
安装好spconv和pcdet后，即可开始模型生成与训练
- Generate the data infos by running the following command:
python -m pcdet.datasets.kitti.kitti_dataset create_kitti_infos tools/cfgs/dataset_configs/kitti_dataset.yaml
会出现TypeError: load() missing 1 required positional argument: ‘Loader’ 的错误，这是由于新的pyyaml 6.0版本不兼容低版本，可以下载5.4.1版本 pip install pyyaml==5.4.1
最终成功的截图如图：
ImportError: libtorch_cpu.so: cannot open shared object file: No such file or directory
删掉build重新编译
assert img_file.exists() AssertionError
/data1/qinjia/OpenPCDet_KITTI/data/kitti/ImageSets中查看测试集、训练集的编码
Error:
WARNING: Imported VTK version (9.1) does not match the one used
to build the TVTK classes (8.1). This may cause problems.
Please rebuild TVTK.
ImportError: cannot import name ‘PILLOW_VERSION’ from ‘PIL’ (/home/fieldrover/anaconda3/envs/PVRCNN_NEW/lib/python3.7/site-packages/PIL/init.py)
Solution:
重装conda install vtk=8.1.2
conda install pillow=8.4
Pytorch错误：Torch not compiled with CUDA enabledpytorch安成了cpu版
检查：
import torch print(torch.__version__) print(torch.cuda.is_available())
重装：
conda install pytorch torchvision cudatoolkit=10.1 -c pytorch
https://www.jianshu.com/p/cacbec7a7b0d
https://blog.csdn.net/u014723479/article/details/103001861
经验：环境问题：删掉rebuild
PVRCNNReference:OpenPCDet_getting_started
The dataset configs are located within tools/cfgs/dataset_configs,
and the model configs are located within
tools/cfgs for different datasets.
如：pv_rcnn.yaml在cfgs/kitti_models/pv_rcnn.yaml
PV-RCNN代码应用——更改测试集文件
PV-RCNN代码应用——参数修改
PV-RCNN代码解读——demo.py & test.py
