search

twni2016 / OrganSegRSTN_PyTorch

PyTorch implementation of OrganSegRSTN - CVPR 2018
http://ccvl.jhu.edu/
MIT License
108 stars 33 forks source link

a weird bug #7

Closed Sungden closed 5 years ago

Sungden commented 5 years ago

When I ran the code, it got back to the error of cuda error:out of memory, although the batch size was 1.

I used the NIH data set and my GPU-memory was 32 G.

Did anybody else have the same issue ever?

Sungden commented 5 years ago

I have addressed this issue by using command 'conda install pytorch=0.4.1 cuda75 -c pytorch' and 'pip install torchvision' to replace command 'conda install -c pytorch pytorch'. To conclude, some command from anaconda.org cannot be used directly. I think using pip comman is more reliable than using conda command in conda virtual environment. So when you find your code which can be run correctly in original server cannot be run in new server, the most possible reason is due to the wrong environment.

Thank the authors for their wonderful job!