In my image classification task, all image areas can provide hints, but some crucial information is right at the edge. Is it a concern? Every conv layer has padding, but I’m unsure if it is difficult for CNNs to extract information right from the image edge. Surely, I could pad the whole input image, but that would add computational and memory cost, too.
1 Like
I wouldn’t think this would be a problem, as long as you’re getting some information from the edges.
However, the best way to find out is to try! Run some experiments and see if you get any difference.