Fine-grained classification on Pets (Adversarial Robustness)
88.3Clean AccuracyCLIP
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| CLIPBackbone=CLIP-ViT-B/16, Perturbation bound (epsilon)=4.02025.12 | 88.3 | 0 | |
| TTPBackbone=CLIP-ViT-B/16, Perturbation bound (epsilon)=4.02025.12 | 88.3 | 64.7 | |
| CLIPBackbone=CLIP-ViT-B/16, Augmented views=15, Perturbation (epsilon)=4.02026.06 | 88.3 | 0 | |
| MTABackbone=CLIP-ViT-B/16, Perturbation bound (epsilon)=4.02025.12 | 88 | 51.8 | |
| TAPTBackbone=CLIP-ViT-B/16, Augmented views=15, Perturbation (epsilon)=4.02026.06 | 87.3 | 14.5 | |
| R-TPTBackbone=CLIP-ViT-B/16, Perturbation bound (epsilon)=4.02025.12 | 87.2 | 60.2 | |
| EnsembleBackbone=CLIP-ViT-B/16, Augmented views=15, Perturbation (epsilon)=4.02026.06 | 86.4 | 28.9 | |
| MTABackbone=CLIP-ViT-B/32, Perturbation Bound (epsilon)=4.02025.12 | 86.3 | 53.6 | |
| EnsembleBackbone=CLIP-ViT-B/16, Perturbation bound (epsilon)=4.02025.12 | 86.2 | 51.2 | |
| SS-TPTBackbone=CLIP-ViT-B/16, Augmented views=15, Perturbation (epsilon)=4.02026.06 | 86.2 | 61.1 | |
| TPTBackbone=CLIP-ViT-B/16, Augmented views=15, Perturbation (epsilon)=4.02026.06 | 86 | 0 | |
| R-TPTBackbone=CLIP-ViT-B/16, Augmented views=15, Perturbation (epsilon)=4.02026.06 | 85.8 | 60.7 | |
| CLIPBackbone=CLIP-ViT-B/32, Perturbation Bound (epsilon)=4.02025.12 | 85.1 | 0 | |
| MTABackbone=CLIP-ResNet50, Epsilon=1.02025.04 | 84.8 | 59.8 | |
| TPTBackbone=CLIP-ResNet50, Epsilon=1.02025.04 | 84.7 | 0.1 | |
| TTPBackbone=CLIP-ViT-B/32, Perturbation Bound (epsilon)=4.02025.12 | 84.7 | 61 | |
| R-TPTBackbone=CLIP-ResNet50, Epsilon=1.02025.04 | 84.6 | 74.2 | |
| R-TPTBackbone=CLIP-ViT-B/32, Perturbation Bound (epsilon)=4.02025.12 | 84.5 | 55.8 | |
| C-TPTBackbone=CLIP-ResNet50, Epsilon=1.02025.04 | 83.6 | 0 | |
| CLIPBackbone=CLIP-ResNet50, Epsilon=1.02025.04 | 83.5 | 0 | |
| TTCBackbone=CLIP-ViT-B/32, Perturbation Bound (epsilon)=4.02025.12 | 83.5 | 11.8 | |
| EnsembleBackbone=CLIP-ResNet50, Epsilon=1.02025.04 | 82.3 | 69.9 | |
| TTCBackbone=CLIP-ViT-B/16, Perturbation bound (epsilon)=4.02025.12 | 82.3 | 10.4 | |
| TTCBackbone=CLIP-ViT-B/16, Augmented views=15, Perturbation (epsilon)=4.02026.06 | 81.1 | 10.8 | |
| DOCBackbone=CLIP-ViT-B/16, Augmented views=15, Perturbation (epsilon)=4.02026.06 | 75.7 | 10.2 | |
| EnsembleBackbone=CLIP-ViT-B/32, Perturbation Bound (epsilon)=4.02025.12 | 75 | 52.5 | |
| CLIPBackbone=CLIP-ViT-B/32, Pre-trained=TeCoA, Augmented views=15, Epsilon=4.02026.06 | 66.9 | 18.5 | |
| TPTBackbone=CLIP-ViT-B/32, Pre-trained=TeCoA, Augmented views=15, Epsilon=4.02026.06 | 64.4 | 27 | |
| SS-TPTBackbone=CLIP-ViT-B/32, Pre-trained=TeCoA, Augmented views=15, Epsilon=4.02026.06 | 60.6 | 38.9 | |
| R-TPTBackbone=CLIP-ViT-B/32, Pre-trained=TeCoA, Augmented views=15, Epsilon=4.02026.06 | 60.4 | 37 | |
| EnsembleBackbone=CLIP-ViT-B/32, Pre-trained=TeCoA, Augmented views=15, Epsilon=4.02026.06 | 59.9 | 37.9 |