Question about the pictures' output resolution.

mit-han-lab / data-efficient-gans

[NeurIPS 2020] Differentiable Augmentation for Data-Efficient GAN Training

https://arxiv.org/abs/2006.10738

BSD 2-Clause "Simplified" License

1.27k stars 175 forks source link

Question about the pictures' output resolution. #52

Open Brand4 opened 3 years ago

Brand4 commented 3 years ago

Hello. In the same dataset created by myself, it can create true pictures when the output resolution is 64×64. However, when the resolution is 128×128 or higher，the resulting pictures are exactly the same except for the color. What may be the cause of this phenomenon or how to get the true pictures with higher resolution? Thanks!

111

222

zsyzzsoft commented 3 years ago

The discriminator may be more easily overfitted when the resolution is higher. You may try to make the augmentations stronger or to stop training earlier.