Image Classification Robustness on Cifar100 (CW, PGD, Auto Attack)
60.45Accuracy (Clean)CLIP
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| CLIP2025.12 | 60.45 | 3.07 | 3.34 | 1.11 | |
| UCATFine-tuning dataset=TinyImageNet, Adversarial training=2-step PGD, Epsilon=1/2552025.12 | 54.45 | 25.16 | 23.92 | 23.03 | |
| FAREFine-tuning dataset=TinyImageNet, Adversarial training=2-step PGD, Epsilon=1/2552025.12 | 48.27 | 2.8 | 2.96 | 0.21 | |
| TGA-ZSRFine-tuning dataset=TinyImageNet, Adversarial training=2-step PGD, Epsilon=1/2552025.12 | 47.37 | 22.82 | 22.64 | 21.75 | |
| PMG-AFTFine-tuning dataset=TinyImageNet, Adversarial training=2-step PGD, Epsilon=1/2552025.12 | 43.59 | 27.23 | 21.31 | 21.18 | |
| TeCoAFine-tuning dataset=TinyImageNet, Adversarial training=2-step PGD, Epsilon=1/2552025.12 | 38.26 | 21.64 | 20.76 | 20.45 |