Image Classification on ImageNet V2 (Clean and Adversarial Accuracy)
63.75Top-1 Accuracy (Clean)Vanilla FT
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Vanilla FTBackbone=ViT-B/322026.03 | 63.75 | 0 | |
| TPGMBackbone=ViT-B/322026.03 | 63.2 | 0 | |
| SPDBackbone=ViT-B/322026.03 | 62.6 | 0 | |
| GRACEBackbone=ViT-B/322026.03 | 62.6 | 20.51 | |
| FLYPBackbone=ViT-B/322026.03 | 62.22 | 0 | |
| WiSE-FTBackbone=ViT-B/322026.03 | 61.06 | 0 | |
| CLIPBackbone=ViT-B/322026.03 | 56.48 | 0 | |
| A-TPTBackbone=ResNet502026.05 | 55.1 | 42.1 | |
| R-TPTBackbone=ResNet502026.05 | 54.9 | 41.6 | |
| MTABackbone=ResNet502026.05 | 54.2 | 24.6 | |
| TPT-EnsembleBackbone=ResNet502026.05 | 52 | 37.2 | |
| CLIPBackbone=ResNet502026.05 | 51.5 | 0.1 | |
| PMG-AFTBackbone=ViT-B/322026.03 | 43.5 | 19.7 | |
| LAATBackbone=ViT-B/322026.03 | 42.3 | 15.8 | |
| TeCoABackbone=ViT-B/322026.03 | 41.1 | 17.71 | |
| FAREBackbone=ViT-B/322026.03 | 39 | 9.77 |