Assume I trained a NN based on 32x32 input images. I want to pass an image with size 512x512 to the trained NN and I would like the trained NN to move convolutionally over the 512x512 images and for each 32x32 patch, the NN will produce an output. The current behavior is that the NN will move convolutionally over the image, but it will produce one output. Is there a way to modify the default behavior of the NN when passing a testing image?
You can just loop over the 512x512 image to get each 32x32 ‘image’ and classify each image. That’s probably the easiest way.