Image Classification on Pets (Acc, Rob)
89.2AccuracyCLBP
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| CLBPSetting=Zero-shot, Backbone=ViT-B/16, Attack=PGD-100, Perturbation Budget (epsilon_a)=4/255, Patch Granularity=16x162026.05 | 89.2 | 79.7 | |
| CLIPSetting=Zero-shot, Backbone=ViT-B/16, Attack=PGD-100, Perturbation Budget (epsilon_a)=4/255, Patch Granularity=16x162026.05 | 88.3 | 0 | |
| C-TPTSetting=Zero-shot, Backbone=ViT-B/16, Attack=PGD-100, Perturbation Budget (epsilon_a)=4/255, Patch Granularity=16x162026.05 | 88.2 | 0 | |
| MTASetting=Zero-shot, Backbone=ViT-B/16, Attack=PGD-100, Perturbation Budget (epsilon_a)=4/255, Patch Granularity=16x162026.05 | 88 | 51.8 | |
| TPTSetting=Zero-shot, Backbone=ViT-B/16, Attack=PGD-100, Perturbation Budget (epsilon_a)=4/255, Patch Granularity=16x162026.05 | 87.4 | 0 | |
| R-TPTSetting=Zero-shot, Backbone=ViT-B/16, Attack=PGD-100, Perturbation Budget (epsilon_a)=4/255, Patch Granularity=16x162026.05 | 87.2 | 60.2 | |
| CLIPBackbone=CLIP-ResNet50, Augmented Views=63, Adversarial perturbation budget (epsilon)=1.02026.06 | 83.6 | 0 | |
| SS-TPTBackbone=CLIP-ResNet50, Augmented Views=63, Adversarial perturbation budget (epsilon)=1.0, mode=test-time2026.06 | 83 | 73.7 | |
| APT+TeCoABackbone=CLIP-ResNet50, Augmented Views=63, Adversarial perturbation budget (epsilon)=1.02026.06 | 79.3 | 79 | |
| TeCoABackbone=CLIP-ResNet50, Augmented Views=63, Adversarial perturbation budget (epsilon)=1.02026.06 | 76 | 75.8 | |
| APTBackbone=CLIP-ResNet50, Augmented Views=63, Adversarial perturbation budget (epsilon)=1.02026.06 | 31.9 | 3.8 |