error testing O-CNN under pytorch after building
opened this issue · comments
Hi, I'm building O-CNN on a docker container with cuda 11.3, Python 3.8.8, pytorch-quantization 2.1.0, pytorch-transformers 1.1.0, torchvision 0.9.0a0, and cudatoolkit 10.2.89 installed. Building works well, but when I run the test cases, I get the following errors. It seems something is wrong when importing nn. How can I solve this error? Thank you so much in advance!
--
$ python -W ignore test/test_all.py -v
Traceback (most recent call last):
File "test/test_all.py", line 3, in
from test_octree_conv import OctreeConvTest
File "/workspace/lidar/windows-doors/O-CNN/pytorch/test/test_octree_conv.py", line 3, in
import ocnn
File "/opt/conda/lib/python3.8/site-packages/ocnn-1.0-py3.8-linux-x86_64.egg/ocnn/init.py", line 4, in
from . import nn
ImportError: /opt/conda/lib/python3.8/site-packages/ocnn-1.0-py3.8-linux-x86_64.egg/ocnn/nn.cpython-38-x86_64-linux-gnu.so: undefined symbol: _ZNK10PointsInfo12check_formatERNSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEEE
According to the error message, I think it is possibly caused by the incompatible gcc version.
What is the Ubuntu version? And please run the following command in the container gcc --version and post the output so that I could diagnose the issue.
The Ubuntu version is Ubuntu 20.04.3, gcc is 9.3.0.
Thank you!
Sorry, I can not reproduce your error.
Please try the official docker provided by pytorch: docker pull pytorch/pytorch:1.9.0-cuda10.2-cudnn7-devel
And here is the ocnn docker I built based on the official pytorch docker: docker pull wangps/ocnn:pytorch-1.9.0