Image Classification on Pets (Clean and Adversarial Top-1 Accuracy)
93.7Clean Top-1 AccuracyMTA
Evaluation results
| Method | Links | ||
|---|---|---|---|
| MTABackbone=ViT-L/14, Perturbation budget (epsilon)=4/2552026.06 | 93.7 | 64.9 | |
| R-TPTBackbone=ViT-L/14, Perturbation budget (epsilon)=4/2552026.06 | 93.7 | 72.9 | |
| T-VSSBackbone=ViT-L/14, Perturbation budget (epsilon)=4/2552026.06 | 93.7 | 73.8 | |
| EnsembleBackbone=ViT-L/14, Perturbation budget (epsilon)=4/2552026.06 | 93.4 | 63.5 | |
| CLIPBackbone=ViT-L/14, Perturbation budget (epsilon)=4/2552026.06 | 93.1 | 0 | |
| TTPBackbone=ViT-L/14, Perturbation budget (epsilon)=4/2552026.06 | 93.1 | 76.3 | |
| TTCBackbone=ViT-L/14, Perturbation budget (epsilon)=4/2552026.06 | 92.2 | 7.6 | |
| CLIPBackbone=ViT-B/16, Perturbation budget (epsilon)=4/2552026.06 | 88.3 | 0 | |
| TTPBackbone=ViT-B/16, Perturbation budget (epsilon)=4/2552026.06 | 88.3 | 64.7 | |
| MTABackbone=ViT-B/16, Perturbation budget (epsilon)=4/2552026.06 | 88 | 51.8 | |
| T-VSSBackbone=ViT-B/16, Perturbation budget (epsilon)=4/2552026.06 | 87.3 | 64.9 | |
| R-TPTBackbone=ViT-B/16, Perturbation budget (epsilon)=4/2552026.06 | 87.2 | 60.2 | |
| EnsembleBackbone=ViT-B/16, Perturbation budget (epsilon)=4/2552026.06 | 86.2 | 51.2 | |
| MTABackbone=ResNet-50, Perturbation budget (epsilon)=1/2552026.06 | 84.8 | 59.8 | |
| TPTBackbone=ResNet-50, Perturbation budget (epsilon)=1/2552026.06 | 84.7 | 0.1 | |
| R-TPTBackbone=ResNet-50, Perturbation budget (epsilon)=1/2552026.06 | 84.6 | 74.2 | |
| T-VSSBackbone=ResNet-50, Perturbation budget (epsilon)=1/2552026.06 | 84.4 | 75.5 | |
| TTPBackbone=ResNet-50, Perturbation budget (epsilon)=1/2552026.06 | 83.8 | 46.9 | |
| C-TPTBackbone=ResNet-50, Perturbation budget (epsilon)=1/2552026.06 | 83.6 | 0 | |
| CLIPBackbone=ResNet-50, Perturbation budget (epsilon)=1/2552026.06 | 83.5 | 0 | |
| EnsembleBackbone=ResNet-50, Perturbation budget (epsilon)=1/2552026.06 | 82.3 | 69.9 | |
| TTCBackbone=ViT-B/16, Perturbation budget (epsilon)=4/2552026.06 | 82.3 | 10.4 |