Adversarial Robustness on CIFAR-100
98.5Final Auto-Attack AccuracyCutout
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Cutout2021.12 | 98.5 | — | — | — | |
| Auto Augment2021.12 | 98.1 | — | — | — | |
| Mixup2021.12 | 97.4 | — | — | — | |
| Outlier Exposure2021.12 | 97.2 | — | — | — | |
| CutMix2021.12 | 97 | — | — | — | |
| Baseline2021.12 | 96.8 | — | — | — | |
| AugMix2021.12 | 95.6 | — | — | — | |
| PIXMIX2021.12 | 92.9 | — | — | — | |
| Sub-ATNorm=l_22021.11 | 37.84 | 38.14 | 0.3 | — | |
| Better DMGenerate data=1M, Epoch=400, Time=≈ 1,438 h2025.05 | 35.65 | — | — | 68.06 | |
| EB-JDAT-SADAJEMGenerate data=n/a, Epoch=100, Time=66.70 h2025.05 | 35.63 | — | — | 68.35 | |
| EB-JDAT-JEM++Generate data=n/a, Epoch=100, Time=31.66 h2025.05 | 35.21 | — | — | 68.09 | |
| ATNorm=l_22021.11 | 32.7 | 36.91 | 4.21 | — | |
| SCOREGenerate data=1M, Epoch=400, Time=≈ 1,438 h2025.05 | 31.4 | — | — | 62.08 | |
| Sub-ATNorm=l_inf2021.11 | 23.83 | 23.89 | 0.06 | — | |
| ATNorm=l_inf2021.11 | 18.12 | 22.83 | 4.71 | — |