alt_cuda_corr backward

[wonozlo]

Hi,

Thanks for such a good research and well organized code!
I have one question about alt_cuda_corr implementation.
In the code, alt_cuda_corr is used as alt_cuda_corr.forward(fmap1,fmap2,coords,radius)
However, backward call is not used in the code.
I saw that multi gpu run also reports unused parameter, which implies that gradient is not fully propagated to the model.
Please tell me if there is something I missed!

Thanks

[XiaoyuShi97]

same question 👍

[eugenelyj]

@wonozlo I think the alt_cuda_corr is only an efficient implementation that are used for testing memory occupation when processing high-resolution input and the authors did not use this implementation for training so the backward function of alt_cuda is never called. By the way, i remember the backward implementation is wrong.

[mazeyu]

I don't how the bug happened but I fixed this in this PR: #186