import error issue

[folkaholic]

when I have prepared the data and the environment，and executed the code： python main.py --config-file configs/sgg_res101_joint.yaml --algorithm grcnn

a error occurred: ImportError: ../graph-rcnn.pytorch/lib/scene_parser/rcnn/_C.cpython-36m-x86_64-linux-gnu.so: undefined symbol: _ZN2at19UndefinedTensorImpl10_singletonE

Is there version incompatibility with my envrionment or other problems？

ps. python=3.6 torch=1.0 cuda=9.0

Thanks!

[lytning98]

I met similar error and solved it by rebuild rcnn module in environment like

pytorch=1.1.0
torchvision=0.3.0
cuda=9.0

python setup.py clean --all
python setup.py build develop

I'm not sure what's the problem before, but cudatoolkit in my previous environment seem to have unmatched CUDA version. Try checking Build column of torch&torchvision in conda list or by other methods if you're not using Conda.

[folkaholic]

I met similar error and solved it by rebuild rcnn module in environment like

pytorch=1.1.0
torchvision=0.3.0
cuda=9.0

python setup.py clean --all
python setup.py build develop

I'm not sure what's the problem before, but cudatoolkit in my previous environment seem to have unmatched CUDA version. Try checking Build column of torch&torchvision in conda list or by other methods if you're not using Conda.

Thank you very much! I have solved this problem yesterday. My cuda version is 9.0,but my torch's cuda version is 10, so I found a suitable match in https://pytorch.org/get-started/previous-versions/ conda install pytorch==1.0.0 torchvision==0.2.1 cuda90 -c pytorch

And the problem solved! Hope it's helpful to other people:) Thank you again!

[lytning98]

Good to know! Btw, how did you prepare dataset with CUDA 9.0? I found that following the instructions I have to run code that supports only tensorflow0.1x based on CUDA8 (for generating RoI distribution)... Or did you just download all files except IMDB? Thanks a lot!

[folkaholic]

Good to know! Btw, how did you prepare dataset with CUDA 9.0? I found that following the instructions I have to run code that supports only tensorflow0.1x based on CUDA8 (for generating RoI distribution)... Or did you just download all files except IMDB? Thanks a lot!

I just downloaded all files except IMDB.

[leeorn]

Hey, I had a similar issue. I Followed the suggestions above (to clean; run conda install pytorch==1.0.0 torchvision==0.2.1 cuda90 -c pytorch; and the build again) But I get the following error

cc1plus: warning: command line option ‘-Wstrict-prototypes’ is valid for C/ObjC but not for C++ cc1plus: error: /usr/local/cuda/lib64/libcudnn.so.7.6.5/include: Not a directory error: command 'gcc' failed with exit status 1

Has anyone encountered that? I'm running in a condo env, as mention, I ran the command above, so I believe that the cuda is 9.0 (although when I run "nvidia-smi" I see "CUDA Version: 10.2").

[wtliao]

Hey, I had a similar issue. I Followed the suggestions above (to clean; run conda install pytorch==1.0.0 torchvision==0.2.1 cuda90 -c pytorch; and the build again) But I get the following error

cc1plus: warning: command line option ‘-Wstrict-prototypes’ is valid for C/ObjC but not for C++ cc1plus: error: /usr/local/cuda/lib64/libcudnn.so.7.6.5/include: Not a directory error: command 'gcc' failed with exit status 1

Has anyone encountered that? I'm running in a condo env, as mention, I ran the command above, so I believe that the cuda is 9.0 (although when I run "nvidia-smi" I see "CUDA Version: 10.2").

Hi Have you solved it? I have the same issue as yours