flip kernel when do the convolution. about dmbsr HOT 4 CLOSED

Hello,

In PyTorch, the function torch.nn.functional.conv2d() implements a cross correlation between the input and the kernel not a convolution as the name would suggest. To implement a convolution, one just need to flip the blur kernel.

I also need to have a cross-correlation function in the code since I need both the operator H (a convolution) and its transpose $H^T$ (a cross correlation) to implement the model.

from dmbsr.

then the main difference between cross-correlation and convolution is flip and not flip right?

from dmbsr.

Yes, the operation is the same but the filter is flipped.

from dmbsr.

thanks!

from dmbsr.