Image Classification on TinyImageNet (Adversarial Robustness Suite)
76.6Clean AccuracyTGA-ZSR
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| TGA-ZSRFine-tuning dataset=TinyImageNet, Adversarial training=2-step PGD, Epsilon=1/2552025.12 | 76.6 | 49.26 | 50.8 | 50.68 | |
| UCATFine-tuning dataset=TinyImageNet, Adversarial training=2-step PGD, Epsilon=1/255, Evaluation protocol=Zero-shot2025.12 | 74.46 | 45.8 | — | — | |
| UCATFine-tuning dataset=TinyImageNet, Adversarial training=2-step PGD, Epsilon=1/2552025.12 | 74.46 | 45.8 | 47.08 | 47.56 | |
| ACATFine-tuning dataset=TinyImageNet, Adversarial training=2-step PGD, Epsilon=1/255, Evaluation protocol=Zero-shot2025.12 | 72.8 | 51.85 | — | — | |
| TeCoAFine-tuning dataset=TinyImageNet, Adversarial training=2-step PGD, Epsilon=1/2552025.12 | 71.24 | 49.44 | 50.16 | 50.96 | |
| DKLFine-tuning dataset=TinyImageNet, Adversarial training=2-step PGD, Epsilon=1/255, Evaluation protocol=Zero-shot2025.12 | 70.84 | 54.71 | — | — | |
| TGA-ZSRTraining Dataset=TinyImageNet, Adversarial Training=10-step PGD, Training Epsilon=2/255, Evaluation Protocol=Zero-shot2025.12 | 69.78 | 11.28 | 29.68 | 30.74 | |
| TRADESFine-tuning dataset=TinyImageNet, Adversarial training=2-step PGD, Epsilon=1/255, Evaluation protocol=Zero-shot2025.12 | 67.73 | 52.19 | — | — | |
| UCATTraining Dataset=TinyImageNet, Adversarial Training=10-step PGD, Training Epsilon=2/255, Evaluation Protocol=Zero-shot2025.12 | 67.18 | 32.84 | 34.64 | 35.38 | |
| TeCoATraining Dataset=TinyImageNet, Adversarial Training=10-step PGD, Training Epsilon=2/255, Evaluation Protocol=Zero-shot2025.12 | 63.2 | 32.68 | 33.9 | 35.74 | |
| CLIPFine-tuning dataset=TinyImageNet, Adversarial training=2-step PGD, Epsilon=1/255, Evaluation protocol=Zero-shot2025.12 | 57.96 | 1.26 | — | — | |
| CLIP2025.12 | 57.96 | 0 | 0.14 | 0.19 | |
| CLIPTraining Dataset=ImageNet-21K / Pre-trained, Evaluation Protocol=Zero-shot2025.12 | 57.96 | 0 | 0 | 0 | |
| PMG-AFTFine-tuning dataset=TinyImageNet, Adversarial training=2-step PGD, Epsilon=1/2552025.12 | 48.6 | 8.22 | 13.16 | 19.18 | |
| FAREFine-tuning dataset=TinyImageNet, Adversarial training=2-step PGD, Epsilon=1/2552025.12 | 41.86 | 0.12 | 4.1 | 3.78 | |
| PMG-AFTTraining Dataset=TinyImageNet, Adversarial Training=10-step PGD, Training Epsilon=2/255, Evaluation Protocol=Zero-shot2025.12 | 22.16 | 0 | 0.02 | 0.12 | |
| FARETraining Dataset=TinyImageNet, Adversarial Training=10-step PGD, Training Epsilon=2/255, Evaluation Protocol=Zero-shot2025.12 | 16.92 | 7 | 7.04 | 8.62 |