Wok

> What is better? Change [32] to [96] or [32,64]? What is the different? I think it should be a power of 2, so 96 would not be valid. https://github.com/lucidrains/lightweight-gan/blob/845eb9d872e020c3c083e1d966811e52d8b26229/lightweight_gan/lightweight_gan.py#L396-L401

Implementation of GSA in the code is from: - https://github.com/lucidrains/global-self-attention-network - https://openreview.net/forum?id=KiFeuZu24k (rejected by ICLR 2021 by the way)  Based on lucidrains' repository, one could refer to this for...

Yes.

The augmentation in the papers mentioned above should not leak. If you experience leaks with the implementation, maybe there is an issue with the implementation or the parameters. See: https://github.com/lucidrains/stylegan2-pytorch#low-amounts-of-training-data...

> Any idea what I'm doing wrong? - d2ed25b36ae228c24a45083df6a9c04f7baf45fe

Whatever fits on your GPU.

It must be one of these. 

People use LPIPS in papers. https://github.com/richzhang/PerceptualSimilarity

> How is this different from FID ? Similar concept, but: - FID measures distance between two distributions of images, - while LPIPS measures distances between two images.

> I can see most of the input images are identical to the output Just chiming in. I trained the model with less than ~200 images (256x256 resolution) recently and...