Classification on OxfordPets
89.4AccuracyProposed
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| ProposedShots=16, Venue=-2025.12 | 89.4 | — | — | — | |
| CLIPShots=0, Venue=ICML’222025.12 | 85.8 | — | — | — | |
| CLIP-PGSImage Encoder=ViT-B/16, p=0.3, Protocol=Zero-shot2025.03 | 58.1 | — | — | — | |
| CLIP-PGSImage Encoder=ViT-B/16, p=0.5, Protocol=Zero-shot2025.03 | 55.7 | — | — | — | |
| E-CLIPImage Encoder=ViT-B/16, Protocol=Zero-shot2025.03 | 55.4 | — | — | — | |
| CLIP-PGSImage Encoder=ViT-S/16, p=0.3, Protocol=Zero-shot2025.03 | 54.5 | — | — | — | |
| CLIPImage Encoder=ViT-B/16, Protocol=Zero-shot2025.03 | 53.9 | — | — | — | |
| A-CLIPImage Encoder=ViT-B/16, Protocol=Zero-shot2025.03 | 51.2 | — | — | — | |
| CLIP-PGSImage Encoder=ViT-S/16, p=0.5, Protocol=Zero-shot2025.03 | 50.6 | — | — | — | |
| FLIPImage Encoder=ViT-B/16, Protocol=Zero-shot2025.03 | 46 | — | — | — | |
| Anti-advMode=Test-time Defence, Attack=10-step PGD, epsilon=1/2552025.03 | — | 20.42 | 80.62 | — | |
| CLIPAttack=10-step PGD, epsilon=1/2552025.03 | — | 1.04 | 87.44 | — | |
| CLIP-FTMode=Adversarial Finetuning, Attack=10-step PGD, epsilon=1/2552025.03 | — | 2.1 | 84.14 | — | |
| FAREMode=Adversarial Finetuning, Attack=10-step PGD, epsilon=1/2552025.03 | — | 31.07 | 79.37 | — | |
| HDMode=Test-time Defence, Attack=10-step PGD, epsilon=1/2552025.03 | — | 12.04 | 80.91 | — | |
| PMG-AFTMode=Adversarial Finetuning, Attack=10-step PGD, epsilon=1/2552025.03 | — | 41.18 | 65.88 | — | |
| RNMode=Test-time Defence, Attack=10-step PGD, epsilon=1/2552025.03 | — | 1.86 | 87.41 | — | |
| TeCoAMode=Adversarial Finetuning, Attack=10-step PGD, epsilon=1/2552025.03 | — | 38.35 | 62.12 | — | |
| TTCMode=Test-time Defence, Attack=10-step PGD, epsilon=1/2552025.03 | — | 57.87 | 83.35 | 56.83 | |
| TTEMode=Test-time Defence, Attack=10-step PGD, epsilon=1/2552025.03 | — | 50.33 | 88.13 | — |