Question about augmentations not using Torchvision

NVlabs / stylegan2-ada-pytorch

StyleGAN2-ADA - Official PyTorch implementation

https://arxiv.org/abs/2006.06676

Other

4.12k stars 1.17k forks source link

Question about augmentations not using Torchvision #107

Open torridgristle opened 3 years ago

torridgristle commented 3 years ago

I'm tackling augmentation for a discriminator's input to improve image generation and noticed StyleGAN2-ADA PyTorch doesn't use Torchvision's transforms for augmentation. For a new project, is there any benefit to using the augmentations here over Torchvision's similar transforms?

kampelmuehler commented 3 years ago

Well in the paper they state that they implemented the augmentations to be differentiable, such that they can train the generator using the augmentations.