Will larger kernel help upsample with better detail or a small sized kernel? And how would padding fit in this scenario?
