Image Classification on ImageNet-A (Adversarial Robustness)
32.69Top-1 Accuracy (Clean)CLIP
Evaluation results
| Method | Links | ||
|---|---|---|---|
| CLIPBackbone=ViT-B/322026.03 | 32.69 | 0 | |
| WiSE-FTBackbone=ViT-B/322026.03 | 32.6 | 0 | |
| TPGMBackbone=ViT-B/322026.03 | 32.38 | 0 | |
| SPDBackbone=ViT-B/322026.03 | 32.07 | 0 | |
| A-TPTBackbone=ResNet502026.05 | 31.6 | 15.3 | |
| FLYPBackbone=ViT-B/322026.03 | 30.06 | 0 | |
| R-TPTBackbone=ResNet502026.05 | 28.4 | 14.4 | |
| MTABackbone=ResNet502026.05 | 27.5 | 5.6 | |
| TPT-EnsembleBackbone=ResNet502026.05 | 22.6 | 10.1 | |
| Vanilla FTBackbone=ViT-B/322026.03 | 22.47 | 0 | |
| CLIPBackbone=ResNet502026.05 | 21.8 | 0 | |
| GRACEBackbone=ViT-B/322026.03 | 20.02 | 3.32 | |
| LAATBackbone=ViT-B/322026.03 | 12.23 | 1.93 | |
| PMG-AFTBackbone=ViT-B/322026.03 | 8.22 | 1.12 | |
| FAREBackbone=ViT-B/322026.03 | 5.97 | 0.54 | |
| TeCoABackbone=ViT-B/322026.03 | 5.11 | 0.66 |