Image Classification on CIFAR-10 (test) (Robustness via AutoAttack)
85.3Accuracy (Clean)FSGM w/ APR-P
Evaluation results
| Method | Links | ||
|---|---|---|---|
| FSGM w/ APR-PBackbone=ResNet-18, Adversarial Training Protocol=FSGM, Augmentation=APR-P2021.08 | 85.3 | 44.1 | |
| FSGM w/ APR-SPBackbone=ResNet-18, Adversarial Training Protocol=FSGM, Augmentation=APR-SP2021.08 | 84.3 | 45.7 | |
| FSGM w/ APR-SBackbone=ResNet-18, Adversarial Training Protocol=FSGM, Augmentation=APR-S2021.08 | 83.5 | 45 | |
| FSGMBackbone=ResNet-18, Adversarial Training Protocol=FSGM2021.08 | 83.3 | 43.2 | |
| FSGM w/ CutoutBackbone=ResNet-18, Adversarial Training Protocol=FSGM, Augmentation=Cutout2021.08 | 81.3 | 41.6 |