Bug: spatialdepthwiseconvolution has memory leak problem

[elliothe]

When I include the spatialdepthwiseconvolution in the network, the GPU memory usage keep growing during training, then running out of memory.

[keillernogueira]

Did you find a solution?

[elliothe]

@keillernogueira , no I tried to migrate the backend function from pytorch to torch, but it looks like taking too much time. Then, I just move to pytorch. They are not maintaining torch anymore, so I suggest you just change the platform. You can check this:
https://github.com/pytorch/pytorch/issues/1708