ncsnv2
ncsnv2 copied to clipboard
rectangular data
Hi there thanks for providing your repo!
I have been trying to adapt the Ncsnv2 model to work with rectangular data, say 5x1024 instead of normal images. I have found that the only noncompatible section is the ConvMeanPool
layer uses the summation:
output = sum([output[:, :, ::2, ::2], output[:, :, 1::2, ::2],
output[:, :, ::2, 1::2], output[:, :, 1::2, 1::2]]) / 4.
Which is the only section that requires square data with 2 even dimensions. Are there any recommendations for some way around this?