Image Classification on ImageNet (val) (Specialized Adversarial Robustness)
63.9Clean AccuracyCLIP
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| CLIPPerturbation budget=4/255, Evaluation protocol=Zero-shot2026.01 | 63.9 | 0 | 0 | 0 | 0 | 0 | 0.1 | 0 | |
| CSRPerturbation budget=4/255, Protocol=Test-Time Defense2026.01 | 62.5 | 60.3 | 60.1 | 46.9 | 46.4 | 45 | 49.7 | 47.6 | |
| FAREPerturbation budget=4/255, Protocol=Adversarial Fine-tuning2026.01 | 48 | 46.2 | 45.9 | 46.7 | 46.3 | 46.3 | 46.6 | 46 | |
| TTCPerturbation budget=4/255, Protocol=Test-Time Defense2026.01 | 40.9 | 2.7 | 0.3 | 25.4 | 6.8 | 6.2 | 10.2 | 1.6 |