Image Classification on STL10 (Clean, AutoAttack, PGD Robustness)
97.03Clean AccuracyCLIP
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| CLIPTraining Data=ImageNet-1k, Adversarial Training=2-step PGD, Epsilon=1/2552025.12 | 97.03 | 35.4 | 3.18 | |
| FARETraining Data=ImageNet-1k, Adversarial Training=2-step PGD, Epsilon=1/2552025.12 | 95.69 | 80.88 | 80.6 | |
| UCATTraining Data=ImageNet-1k, Adversarial Training=2-step PGD, Epsilon=1/2552025.12 | 95.65 | 82.09 | 81.73 | |
| TeCoATraining Data=ImageNet-1k, Adversarial Training=2-step PGD, Epsilon=1/2552025.12 | 93.3 | 83.45 | 83.16 |