I have a question and hope it's valid. What is the difference between using a convolution and a subsequent pooling layer vs image transformation in the discriminator of a GAN? Can the discriminator for example capture attributes of images in their entirety better when I input 300*300 images using convolution and pooling vs resizing them before putting them in a linear layer? I guess I dont understand it to the point yet and would like to change that! Thank you
Asked
Active
Viewed 19 times
