当我在终端里
python tools/test.py configs/mask_rcnn_x101_64x4d_fpn_1x.py checkpoints/mask_rcnn_x101_64x4d_fpn_1x_20181218-cb159987.pth --out results.pkl --eval bbox segm
就出现
ImportError: /home/bluo/mmdetection/mmdet/ops/dcn/deform_conv_cuda.cpython-36m-x86_64-linux-gnu.so: undefined symbol: _ZN3c105ErrorC1ENS_14SourceLocationERKSs
错误
请问给如何解决?
ubuntu18.04
conda新建的环境环境:Python3.6.8、pytorch1.0.1
Please use English or English&Chinese for issues. Also please use the Error report issue template.
then how to solve it ?
Have you soloved this ? I also meet this, thanks!!!!
When I use anaconda to create a virtual environment of mmdetection, and the settings are following:
First, I have a cuda10.0 in /usr/local/cuda-10.0.
then the mmdetection environment used pytorch1.2 cudatoolkit10.0, but the driver of GPU is 418.
the problem of undefined symbol: _ZN3c105ErrorC1ENS_14SourceLocationERKSs will occur
but I used pytorch1.3, cudatoolkit10.1 and the CUDA_HOME=/usr/local/cuda-10.1.
the code can run.
I guess the reason for the problem is not matching in some librarys 's version.
I meet this problem too. I solve it by installing pytorch from pytorch official channel, instead of mirror such as tsinghua and alibaba.
Most helpful comment
Please use English or English&Chinese for issues. Also please use the
Error reportissue template.