JYChen18
commented
4 months ago Sorry for the late reply.
It is due to the misalignment between the CUDA you use to compile the lib and the CUDA you use to run.
You may need to change the environment variable $CUDA_HOME to point to a system CUDA with version 11.x, and recompile the pointnet2. Or change the pytorch version.
Please also see #5 .
pytorch=1.9.1 +CUDA11.1 无法运行，如何适配？ 编译pointnet2_cuda时报错。 Traceback (most recent call last): File "network/train.py", line 16, in
from datasets.dataset import get_dataloader
File "/hy-tmp/HOTrack-main/network/../datasets/dataset.py", line 15, in
from SimGrasp_dataset import SimGraspDataset
File "/hy-tmp/HOTrack-main/network/../datasets/SimGrasp_dataset.py", line 8, in
from network.models.hand_utils import handkp2palmkp
File "/hy-tmp/HOTrack-main/network/../datasets/../network/models/hand_utils.py", line 3, in
from pointnet_utils import knn_point, group_operation
File "/hy-tmp/HOTrack-main/network/../datasets/../network/models/pointnet_utils.py", line 10, in
from pointnet_lib import pointnet2_utils as futils
File "/hy-tmp/HOTrack-main/network/../datasets/../network/models/pointnet_lib/pointnet2_utils.py", line 7, in
import pointnet2_cuda as pointnet2
ImportError: /usr/local/lib/python3.8/dist-packages/pointnet2-0.0.0-py3.8-linux-x86_64.egg/pointnet2_cuda.cpython-38-x86_64-linux-gnu.so: undefined symbol: _ZN6caffe28TypeMeta21_typeMetaDataInstanceIdEEPKNS_6detail12TypeMetaDataEv