search

facebookresearch / pytorch_GAN_zoo

A mix of GAN implementations including progressive growing
BSD 3-Clause "New" or "Revised" License
1.61k stars 270 forks source link

Upscale2d() vs torch.nn.functional.interpolate #80

Open mbahri opened 5 years ago

mbahri commented 5 years ago

Hi

I was simply wondering if there was any particular reason to retain the knn interpolation from the original Progressive GAN code (implemented in the Upscale2d function https://github.com/facebookresearch/pytorch_GAN_zoo/blob/master/models/networks/custom_layers.py) instead of torch.nn.functional.interpolate.

Are there differences between the two that I missed?

Thanks

Molugan commented 4 years ago

Hi,

Sorry for the delay.

GAN are unstable and When I was reproducing NVIDIA's code I had at first some issues getting similar results and thus I kept the original method. But you're right, there is no reason not to use torch.nn.functional.interpolate now, I'll add an issue.