search

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

Used for other condition gan model #77

Closed zhangqianhui closed 3 years ago

zhangqianhui commented 3 years ago

hi

If this method is used for pix2pix, stargan, whether the augmentation are applied for reconstruction or cycle loss or not?

zsyzzsoft commented 3 years ago

Not necessary. But if used, it may stand for a different purpose other than reducing discriminator overfitting.

zhangqianhui commented 3 years ago

Thanks!