Fine-grained classification on Flower102 (Adversarial Evaluation)
76.5Clean AccuracyTTC
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| TTCBackbone=ViT-L/14, Epsilon=4/2552026.05 | 76.5 | 7.5 | |
| CLIPBackbone=ViT-L/14, Epsilon=4/2552026.05 | 76.2 | 0 | |
| R-TPTBackbone=ViT-L/14, Epsilon=4/2552026.05 | 76.2 | 55.6 | |
| MTABackbone=ViT-L/14, Epsilon=4/2552026.05 | 76.1 | 44.2 | |
| TTPBackbone=ViT-L/14, Epsilon=4/2552026.05 | 76.1 | 58.7 | |
| AGCBackbone=ViT-L/14, Epsilon=4/2552026.05 | 76 | 96 | |
| EnsembleBackbone=ViT-L/14, Epsilon=4/2552026.05 | 75 | 60 | |
| TTCBackbone=CLIP-ViT-B/16, Perturbation bound (epsilon)=4.02025.12 | 69 | 7.4 | |
| R-TPTBackbone=CLIP-ViT-B/16, Perturbation bound (epsilon)=4.02025.12 | 68.7 | 44.6 | |
| CLIPBackbone=CLIP-ViT-B/16, Perturbation bound (epsilon)=4.02025.12 | 67.4 | 0 | |
| MTABackbone=CLIP-ViT-B/16, Perturbation bound (epsilon)=4.02025.12 | 67.4 | 27.9 | |
| CLIPBackbone=CLIP-ViT-B/16, Augmented views=15, Perturbation (epsilon)=4.02026.06 | 67.4 | 0 | |
| TPTBackbone=CLIP-ViT-B/16, Augmented views=15, Perturbation (epsilon)=4.02026.06 | 67.4 | 0 | |
| TTPBackbone=CLIP-ViT-B/16, Perturbation bound (epsilon)=4.02025.12 | 67.3 | 47.2 | |
| SS-TPTBackbone=CLIP-ViT-B/16, Augmented views=15, Perturbation (epsilon)=4.02026.06 | 67 | 45.4 | |
| R-TPTBackbone=CLIP-ViT-B/16, Augmented views=15, Perturbation (epsilon)=4.02026.06 | 66.7 | 44.7 | |
| EnsembleBackbone=CLIP-ViT-B/16, Perturbation bound (epsilon)=4.02025.12 | 65.9 | 36.3 | |
| EnsembleBackbone=CLIP-ViT-B/16, Augmented views=15, Perturbation (epsilon)=4.02026.06 | 65.6 | 17.1 | |
| TTCBackbone=CLIP-ViT-B/16, Augmented views=15, Perturbation (epsilon)=4.02026.06 | 65 | 7.5 | |
| C-TPTBackbone=CLIP-ResNet50, Epsilon=1.02025.04 | 64.8 | 0 | |
| DOCBackbone=CLIP-ViT-B/16, Augmented views=15, Perturbation (epsilon)=4.02026.06 | 64.6 | 9.7 | |
| MTABackbone=CLIP-ViT-B/32, Perturbation Bound (epsilon)=4.02025.12 | 64.4 | 36.5 | |
| TTCBackbone=CLIP-ViT-B/32, Perturbation Bound (epsilon)=4.02025.12 | 64.3 | 3.2 | |
| CLIPBackbone=CLIP-ViT-B/32, Perturbation Bound (epsilon)=4.02025.12 | 64 | 0 | |
| TTPBackbone=CLIP-ViT-B/32, Perturbation Bound (epsilon)=4.02025.12 | 63.6 | 42 | |
| TAPTBackbone=CLIP-ViT-B/16, Augmented views=15, Perturbation (epsilon)=4.02026.06 | 63.1 | 9.1 | |
| R-TPTBackbone=CLIP-ViT-B/32, Perturbation Bound (epsilon)=4.02025.12 | 62.6 | 37.6 | |
| TPTBackbone=CLIP-ResNet50, Epsilon=1.02025.04 | 62.1 | 0 | |
| CLIPBackbone=CLIP-ResNet50, Epsilon=1.02025.04 | 61.7 | 0 | |
| MTABackbone=CLIP-ResNet50, Epsilon=1.02025.04 | 61 | 31.5 | |
| R-TPTBackbone=CLIP-ResNet50, Epsilon=1.02025.04 | 60.6 | 51.9 | |
| EnsembleBackbone=CLIP-ViT-B/32, Perturbation Bound (epsilon)=4.02025.12 | 58.1 | 36.1 | |
| EnsembleBackbone=CLIP-ResNet50, Epsilon=1.02025.04 | 58 | 46.6 | |
| CLIPBackbone=CLIP-ViT-B/32, Pre-trained=TeCoA, Augmented views=15, Epsilon=4.02026.06 | 30.8 | 10 | |
| TPTBackbone=CLIP-ViT-B/32, Pre-trained=TeCoA, Augmented views=15, Epsilon=4.02026.06 | 27.7 | 12.5 | |
| EnsembleBackbone=CLIP-ViT-B/32, Pre-trained=TeCoA, Augmented views=15, Epsilon=4.02026.06 | 26.6 | 15.9 | |
| SS-TPTBackbone=CLIP-ViT-B/32, Pre-trained=TeCoA, Augmented views=15, Epsilon=4.02026.06 | 26.3 | 16.2 | |
| R-TPTBackbone=CLIP-ViT-B/32, Pre-trained=TeCoA, Augmented views=15, Epsilon=4.02026.06 | 26 | 15.5 |