using TORCH_CHECK instead of AT_CHECK and adding .contiguous();

[elmahyai]

AT_CHECK is deprecated in future torch versions adding this line solves the problem.

The second problem I encounters was that I get this error in the backward pass of the deform_conv_cuda function.

view size is not compatible with input tensor's size and stride (at least one dimension spans across two contiguous subspaces). Call .contiguous() before .view().

The solution was to add .contiguous() and the error is gone.