Image Generation on CIFAR-10 (train)
7.8FIDReal data
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Real data2018.02 | 7.8 | 11.24 | |
| E2GAN-top1Search Cost (GPU days)=0.32020.07 | 11.26 | — | |
| AutoGAN-top1Search Cost (GPU days)=22020.07 | 12.42 | — | |
| E2GAN-top3Search Cost (GPU days)=0.32020.07 | 12.48 | — | |
| E2GAN-top2Search Cost (GPU days)=0.32020.07 | 12.96 | — | |
| AutoGAN-top2Search Cost (GPU days)=22020.07 | 13.67 | — | |
| AutoGAN-top3Search Cost (GPU days)=22020.07 | 13.68 | — | |
| Improv MMD GANArchitecture=Manually designed2020.07 | 16.21 | 8.29 | |
| Random search-1Search Cost (GPU days)=None2020.07 | 17.34 | 8.09 | |
| Random search-2Search Cost (GPU days)=None2020.07 | 21.39 | 7.97 | |
| SN-GANsBackbone Architecture=ResNet, Optimization variant=Eq.(17)2018.02 | 21.7 | 8.22 | |
| OrthonormalBackbone Architecture=ResNet, Optimization variant=Eq.(17)2018.02 | 23.8 | 7.92 | |
| ProbGANArchitecture=Manually designed2020.07 | 24.6 | 7.75 | |
| SN-GANsBackbone Architecture=Standard CNN, Optimization variant=Eq.(17)2018.02 | 25.5 | 7.58 | |
| MGANArchitecture=Manually designed2020.07 | 26.7 | — | |
| OrthonormalBackbone Architecture=Standard CNN2018.02 | 29 | 7.4 | |
| SN-GANsBackbone Architecture=Standard CNN2018.02 | 29.3 | 7.42 | |
| AGANSearch Cost (GPU days)=12002020.07 | 30.5 | — | |
| Layer Norm.Backbone Architecture=Standard CNN2018.02 | 33.9 | 7.19 | |
| Weight Norm.Backbone Architecture=Standard CNN2018.02 | 34.7 | 6.84 | |
| GAN-GPBackbone Architecture=Standard CNN2018.02 | 37.7 | 6.93 | |
| WGAN-GPBackbone Architecture=Standard CNN2018.02 | 40.2 | 6.68 | |
| Weight clippingBackbone Architecture=Standard CNN2018.02 | 42.6 | 6.41 | |
| Batch Norm.Backbone Architecture=Standard CNN2018.02 | 56.3 | 6.27 | |
| DCGAN2018.02 | — | 6.64 | |
| LR-GANs2018.02 | — | 7.17 | |
| Warde-Farley et al.2018.02 | — | 7.72 | |
| WGAN-GPBackbone Architecture=ResNet2018.02 | — | 7.86 |