Image Generation on FFHQ 1k (train)
1.18FID (SwAV)Vision-aided Adversarial training
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Vision-aided Adversarial trainingAugmentation Strategy=ADA, Fine-tuning=true2021.12 | 1.18 | — | — | |
| Vision-aided Adversarial trainingAugmentation Strategy=DiffAugment, Fine-tuning=true2021.12 | 1.95 | — | — | |
| ADAAugmentation Strategy=ADA2021.12 | 4.15 | — | — | |
| DiffAugmentAugmentation Strategy=DiffAugment2021.12 | 6.31 | — | — | |
| StyleGAN2Augmentation Strategy=none2021.12 | 14.42 | — | — | |
| ADAResolution=256x256, Backbone=StyleGAN22021.11 | — | 22.59 | 4.239 | |
| ADA + APAResolution=256x256, Backbone=StyleGAN22021.11 | — | 18.892 | 4.316 | |
| APAResolution=256x256, Backbone=StyleGAN22021.11 | — | 45.192 | 4.13 | |
| StyleGAN2Resolution=256x256, Backbone=StyleGAN22021.11 | — | 86.407 | 2.806 |