How can i run this project on RTX3090? about cmrnet

[cattaneod]

Are you sure that you are using CUDA 11.1? From the error it seems that CUDA is older than 11.
can you try to run:
/usr/local/cuda/bin/nvcc --version
and paste the output?

[cattaneod]

Can you try only with arch=compute_80, code=sm_80, without the second argument arch=compute_80, code=compute_80

[WJ1214]

Are you sure that you are using CUDA 11.1? From the error it seems that CUDA is older than 11. can you try to run: /usr/local/cuda/bin/nvcc --version and paste the output?

Thank you, I may know the reason why.

[WJ1214]

Can you try only with arch=compute_80, code=sm_80, without the second argument arch=compute_80, code=compute_80

Let me try to change the CUDA version and nvcc_args, I will give you a result.

[WJ1214]

Can you try only with arch=compute_80, code=sm_80, without the second argument arch=compute_80, code=compute_80

Hi! Now I can build project successfully, but I get another problem... It returns this error when I try to run train model script:

It's weird...

[cattaneod]

Now the error is probably that you have installed pytorch with a different version of cuda than the one that you used to compile the packages.

I suggest that you create a new python environment (e.g. conda environment), install pytorch with the correct cuda version that you will use to compile the packages, and finally, before running python setup.py install, remove the build folder with rm -rf build

[WJ1214]

OK, I will try it. Thank you!

[cattaneod]

And, as mentioned in #4, can you modify the file models/CMRNet/correlation_package/correlation.py and invert the first two lines?

[WJ1214]

And, as mentioned in #4, can you modify the file models/CMRNet/correlation_package/correlation.py and invert the first two lines?

You mean that I need to import torch first and then import correlation_cuda in correlation.py?

[cattaneod]

You mean that I need to import torch first and then import correlation_cuda in correlation.py?

Yes, exactly

[WJ1214]

You mean that I need to import torch first and then import correlation_cuda in correlation.py?

Yes, exactly

OKay...I will try it

[WJ1214]

It worked! Thank you very much!

[cattaneod]

Thank you for your feedback, I updated the code to run on newer versions of pytorch and with cuda 11.x