Zero-shot Image Classification on Average Zero-shot (15 datasets)
60.4Clean AccuracyWISE-FT
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| WISE-FTBackbone=ViT-B/32, Pre-training Dataset=ImageNet, Evaluation Protocol=Zero-shot2026.03 | 60.4 | 0 | |
| GRACEBackbone=ViT-B/32, Pre-training Dataset=ImageNet, Evaluation Protocol=Zero-shot2026.03 | 59.61 | 21.1 | |
| CLIPBackbone=ViT-B/32, Evaluation Protocol=Zero-shot2026.03 | 59.31 | 0 | |
| FAREAveraged across 15 datasets=true, Zero-shot evaluation=true, Adversarial Attack=PGD-100 (epsilon = 1/255), Backbone=ViT-B/322026.02 | 58.68 | 17.88 | |
| TPGMBackbone=ViT-B/32, Pre-training Dataset=ImageNet, Evaluation Protocol=Zero-shot2026.03 | 58.23 | 0 | |
| SPDBackbone=ViT-B/32, Pre-training Dataset=ImageNet, Evaluation Protocol=Zero-shot2026.03 | 58.01 | 0 | |
| Vanilla FTBackbone=ViT-B/32, Pre-training Dataset=ImageNet, Evaluation Protocol=Zero-shot2026.03 | 57.21 | 0 | |
| SAFT-MAveraged across 15 datasets=true, Zero-shot evaluation=true, Adversarial Attack=PGD-100 (epsilon = 1/255), Backbone=ViT-B/322026.02 | 56.01 | 33.78 | |
| SAFT-LAveraged across 15 datasets=true, Zero-shot evaluation=true, Adversarial Attack=PGD-100 (epsilon = 1/255), Backbone=ViT-B/322026.02 | 55.27 | 35.35 | |
| TGA-ZSRAveraged across 15 datasets=true, Zero-shot evaluation=true, Adversarial Attack=PGD-100 (epsilon = 1/255), Backbone=ViT-B/322026.02 | 54.58 | 31.5 | |
| FLYPBackbone=ViT-B/32, Pre-training Dataset=ImageNet, Evaluation Protocol=Zero-shot2026.03 | 54.37 | 0 | |
| CLIPAveraged across 15 datasets=true, Zero-shot evaluation=true, Adversarial Attack=PGD-100 (epsilon = 1/255), Backbone=ViT-B/322026.02 | 52.72 | 6.62 | |
| PMG-AFTBackbone=ViT-B/32, Pre-training Dataset=ImageNet, Evaluation Protocol=Zero-shot2026.03 | 45.18 | 22 | |
| FAREBackbone=ViT-B/32, Pre-training Dataset=ImageNet, Evaluation Protocol=Zero-shot2026.03 | 44.66 | 19.01 | |
| LAATBackbone=ViT-B/32, Pre-training Dataset=ImageNet, Evaluation Protocol=Zero-shot2026.03 | 43.46 | 19.58 | |
| TeCoAAveraged across 15 datasets=true, Zero-shot evaluation=true, Adversarial Attack=PGD-100 (epsilon = 1/255), Backbone=ViT-B/322026.02 | 43.08 | 26 | |
| TeCoABackbone=ViT-B/32, Pre-training Dataset=ImageNet, Evaluation Protocol=Zero-shot2026.03 | 38.75 | 20.78 | |
| PMG-AFTAveraged across 15 datasets=true, Zero-shot evaluation=true, Adversarial Attack=PGD-100 (epsilon = 1/255), Backbone=ViT-B/322026.02 | 37.42 | 27.54 |