Image Classification on ImageNet (test) (OOD Robustness Metrics)
82.7ImageNet (ID) AccuracyARF
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| ARFBackbone=ViT-B/16, Protocol=Anchor-based Robust Fine-tuning2024.04 | 82.7 | 72.8 | 75.6 | 50.3 | 51.8 | 55.8 | 61.3 | |
| FLYPBackbone=ViT-B/16, Protocol=FLYP2024.04 | 82.6 | 73 | 71.4 | 48.1 | 49.6 | 54.7 | 59.4 | |
| LP-FTBackbone=ViT-B/16, Protocol=Linear Probing then Fine-tuning2024.04 | 81.7 | 72.1 | 73.5 | 47.6 | 50.3 | 54.4 | 59.6 | |
| FTBackbone=ViT-B/16, Protocol=Fine-tuning2024.04 | 81.3 | 71.2 | 66.1 | 37.8 | 46.1 | 51.6 | 54.6 | |
| LPBackbone=ViT-B/16, Protocol=Linear Probing2024.04 | 79.9 | 69.8 | 70.8 | 46.4 | 46.9 | 50.4 | 56.9 | |
| CLIPBackbone=ViT-B/16, Protocol=Zero-shot2024.04 | 68.3 | 61.9 | 77.7 | 50 | 48.3 | 54.2 | 58.4 |