It seems like these filters as represented in the middle can be useful for detecting features, like edges, shapes, or other spatial patterns in the input image.
lordpancake
A great way to think about filters or convolutions is that they are simply another channel (or dimension) to an image, as in RGB, so it follows that you could have more channels
It seems like these filters as represented in the middle can be useful for detecting features, like edges, shapes, or other spatial patterns in the input image.