Zero-shot Image Classification on DTD (Robustness)
39.1Robust Accuracy (PGD-100, eps=1/255)CSR
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| CSR2026.01 | 39.1 | 41.9 | — | — | — | |
| TAMEAttack Method=AutoAttack, Backbone=ViT-B/16, Input Configuration=V-L Independent2026.05 | 36.8 | — | — | — | — | |
| TAMEAttack Method=PGD, Backbone=ViT-B/16, Input Configuration=V-L Independent2026.05 | 36.3 | — | — | — | — | |
| R-TPT2026.01 | 36.2 | 44.2 | — | — | — | |
| FARE2026.01 | 34.3 | 35.3 | — | — | — | |
| TAMEBackbone=ViT-B/32, Evaluation Protocol=Zero-shot, Prompt Configuration=V-L Independent2026.05 | 31.1 | — | 34 | 35.1 | 30.1 | |
| TeCoA2026.01 | 29.3 | 29.9 | — | — | — | |
| LPF2026.01 | 26.7 | 40.5 | — | — | — | |
| GRACEBackbone=ViT-B/16, Training=ImageNet, Evaluation=Zero-shot2026.03 | 25.3 | 52.6 | — | — | — | |
| PMG-AFTBackbone=ViT-B/16, Training=ImageNet, Evaluation=Zero-shot2026.03 | 22.15 | 34.38 | — | — | — | |
| Anti-Adv2026.01 | 22.1 | 40.6 | — | — | — | |
| TTC2026.01 | 22.1 | 35.7 | — | — | — | |
| TTE2026.01 | 20.6 | 41.9 | — | — | — | |
| LAATBackbone=ViT-B/16, Training=ImageNet, Evaluation=Zero-shot2026.03 | 20.5 | 32.1 | — | — | — | |
| TeCoABackbone=ViT-B/16, Training=ImageNet, Evaluation=Zero-shot2026.03 | 18.16 | 28.93 | — | — | — | |
| FAREBackbone=ViT-B/16, Training=ImageNet, Evaluation=Zero-shot2026.03 | 12.8 | 38.5 | — | — | — | |
| HD2026.01 | 8.5 | 40.4 | — | — | — | |
| VanillaBackbone=ViT-B/32, Evaluation Protocol=Zero-shot2026.05 | 5 | — | 0.9 | 11.1 | 0 | |
| CLIP2026.01 | 0.1 | 42.3 | — | — | — | |
| CLIPBackbone=ViT-B/16, Training=ImageNet, Evaluation=Zero-shot2026.03 | 0 | 43.85 | — | — | — | |
| Vanilla FTBackbone=ViT-B/16, Training=ImageNet, Evaluation=Zero-shot2026.03 | 0 | 52.3 | — | — | — | |
| WISE-FTBackbone=ViT-B/16, Training=ImageNet, Evaluation=Zero-shot2026.03 | 0 | 48.6 | — | — | — | |
| FLYPBackbone=ViT-B/16, Training=ImageNet, Evaluation=Zero-shot2026.03 | 0 | 50.2 | — | — | — | |
| TPGMBackbone=ViT-B/16, Training=ImageNet, Evaluation=Zero-shot2026.03 | 0 | 51.4 | — | — | — | |
| SPDBackbone=ViT-B/16, Training=ImageNet, Evaluation=Zero-shot2026.03 | 0 | 50.8 | — | — | — |