参考:
正文:
OpenPCDet 升级到 v0.5.2 ,支持了更多的数据。
$ conda create -n OpenPCDetv052 python=3.8
$ conda activate OpenPCDetv052
$ cat /usr/local/cuda/version.txt
CUDA Version 10.2.89
$ python -m pip install spconv-cu102
$ conda install pytorch==1.8.1 torchvision==0.9.1 cudatoolkit=10.2 -c pytorch
$ git clone https://github.com/open-mmlab/OpenPCDet.git
$ cd OpenPCDet
$ vim requirements.txt
去掉 torch>=1.1 torchvision 这两行,详细原因后面讲
$ cat requirements.txt
numpy
llvmlite
numba
tensorboardX
easydict
pyyaml
scikit-image
tqdm
SharedArray
$ python -m pip install -r requirements.txt
$ python setup.py develop
$ python
#检验 pytorch
>>> import torch
>>> print(torch.__version__)
1.8.1
>>> torch.cuda.is_available()
True
#检验 spconv
>>> import spconv
#检验 pcdet
>>> import pcdet
>>> exit()
error:
RuntimeError: module compiled against API version 0xf but this version of numpy is 0xe
Traceback (most recent call last):
...
import SharedArray
ImportError: numpy.core.multiarray failed to import
python -m pip install numpy==1.22.0
ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
numba 0.55.1 requires numpy<1.22,>=1.18, but you have numpy 1.22.0 which is incompatible.
后续有问题了再把版本降下来。
运行 python train.py --cfg_file cfgs/kitti_models/pointpillar.yaml 出现这个问题
只要把numpy 更新到合适的版本,试验得出 numpy==1.21.5
python -m pip install numpy==1.21.5
运行 python train.py --cfg_file cfgs/kitti_models/pointpillar.yaml 出现这个问题
查看问题出自哪儿,参考:
动态库链接出现undefined symbol: _ZN16XXXXXXXD1Ev
undefined symbol问题的查找定位与解决方法
symbol lookup error问题问题的查找、定位与解决方法
其实还是 pytorch 版本问题,多次试验得出
pytorch1.8.1
torchvision0.9.1
可以运行 train.py
大功告成!@!