Hi, I am attempting to reimplement a version of CycleGan from pytorch to julia. But I have a doubt about how to use a package from Flux.
Looking at the example code from Pytorch, there is a property in ConvTranspose2d called output_padding, but there is no such thing in Flux.ConvTranspose. Because of this, my output size is not correct.
Another property that I couldn’t find was padding_mode, nor in Conv or in ConvTranspose, is there a way to make a work around? How should I treat this one?
I even try to use Flux.SamePad() to correct this issue, and works for some sizes as (100,100) but (3,3) is not working.
Thanks in advanced.