Grid-Anchor-based-Image-Cropping-Pytorch
Grid-Anchor-based-Image-Cropping-Pytorch copied to clipboard
lld533
segmentation fault
the config： pytorch=1.3.0 torchvision=0.4.1 cuda=10.0 cudatoolkit=10.0
the GPUs are tesla V100-sxm2-16gb
After I modified the make.sh‘s gpu-arch and made it，when running the demo_eval.py，it comes the fault： segmentation fault
(gdb) f 4 #4 push_back (__x=<unknown type in /home/seclogin/anaconda3/envs/gc_torch1py35/lib/python3.5/site-packages/roi_align_api-0.0.0-py3.5-linux-x86_64.egg/roi_align_api.cpython-35m-x86_64-linux-gnu.so, CU 0x0, DIE 0x12a6de>, this=0x558a385893c8) at /usr/include/c++/4.8.2/bits/stl_vector.h:920 920 { emplace_back(std::move(__x)); } (gdb) f 59 #59 0x0000558a35c25571 in main ()
could someone tell me how to use the project rightly？
@lld533 ask for help！！ I cannot use the demo_eval.py to make some test Thanks for your question. There are a few things that I'm not sure. 1/ It seems that you are using early c++ version. In fact, I use version 5. 2/ You are using pytorch 1.3, but I only test my code with 1.0/1.1/1.2 (no test under PyTorch 1.3 or later). 3/ Sorry, we have no sxm2 version GPU card or Tesla V100 (16G/32G, pcie version or sxm2 version). I think the closest GPU we have is Quadro GV100. Of course, this should not be the problem.
At the time being, please 1/ upgrade your compiler, e.g. gcc-5.4 or other 5.x version 2/ downgrade pytorch version to 1.2, and 3/ re-compile the code.
Let me know if it still fails to work.
@lld533 ask for help！！ I cannot use the demo_eval.py to make some test Thanks for your question. There are a few things that I'm not sure. 1/ It seems that you are using early c++ version. In fact, I use version 5. 2/ You are using pytorch 1.3, but I only test my code with 1.0/1.1/1.2 (no test under PyTorch 1.3 or later). 3/ Sorry, we have no sxm2 version GPU card or Tesla V100 (16G/32G, pcie version or sxm2 version). I think the closest GPU we have is Quadro GV100. Of course, this should not be the problem.
At the time being, please 1/ upgrade your compiler, e.g. gcc-5.4 or other 5.x version 2/ downgrade pytorch version to 1.2, and 3/ re-compile the code.
Let me know if it still fails to work.
After I make some try with your advice, it still doesn't work!!
@lld533 ask for help！！ I cannot use the demo_eval.py to make some test Thanks for your question. There are a few things that I'm not sure. 1/ It seems that you are using early c++ version. In fact, I use version 5. 2/ You are using pytorch 1.3, but I only test my code with 1.0/1.1/1.2 (no test under PyTorch 1.3 or later). 3/ Sorry, we have no sxm2 version GPU card or Tesla V100 (16G/32G, pcie version or sxm2 version). I think the closest GPU we have is Quadro GV100. Of course, this should not be the problem.
At the time being, please 1/ upgrade your compiler, e.g. gcc-5.4 or other 5.x version 2/ downgrade pytorch version to 1.2, and 3/ re-compile the code. Let me know if it still fails to work.
After I make some try with your advice, it still doesn't work!!
Hello, could you please double check if images are loaded correctly and make sure the image is not None? I came across similar problem when the test images were not properly loaded. I'll go through my code soon. Thanks.
Hello, could you please double check if images are loaded correctly and make sure the image is not None? I came across similar problem when the test images were not properly loaded. I'll go through my code soon. Thanks.
Look at the upper snip-image, data loader looks like no problem, but when net infers the data, segmentation fault comes across again.