After a series of convolutions, I am up-sampling a compressed representation, I was curious what is the methodology I should follow to choose an optimum kernel size for up-sampling.
How will the filter (or kernel) size affect the transpose convolution operation (e.g. when using
ConvTranspose2d)? Will a larger kernel help upsample with better detail or a small-sized kernel? And how would padding fit in this scenario?
At what rate should the depth decrease while upsampling i.e from (Dx24x24) to (D/2 or D/4, 48, 48)