Image Classification on DTD (Adversarial Robustness Metrics)
40.64Clean AccuracyCLIP
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| CLIPTraining Data=ImageNet-1k, Adversarial Training=2-step PGD, Epsilon=1/2552025.12 | 40.64 | 2.5 | 0.0019 | |
| UCATTraining Data=ImageNet-1k, Adversarial Training=2-step PGD, Epsilon=1/2552025.12 | 37.18 | 24.42 | 0.2245 | |
| FARETraining Data=ImageNet-1k, Adversarial Training=2-step PGD, Epsilon=1/2552025.12 | 36.86 | 24.31 | 0.2202 | |
| TeCoATraining Data=ImageNet-1k, Adversarial Training=2-step PGD, Epsilon=1/2552025.12 | 34.15 | 22.88 | 0.207 |