Image Classification on DTD (Clean, AutoAttack)
32.34AutoAttack RobustnessHPT-GPD
Evaluation results
| Method | Links | ||
|---|---|---|---|
| HPT-GPDBackbone=ViT-B/32, Proxy CLIP=ViT-B/16, Attack Protocol=PGD-10, Training time per epoch (s)=14862026.01 | 32.34 | 34.26 | |
| PMGBackbone=ViT-B/32, Proxy CLIP=ViT-B/16, Attack Protocol=PGD-10, Training time per epoch (s)=18172026.01 | 29.63 | 35.11 | |
| FT-TeCoABackbone=ViT-B/32, Proxy CLIP=ViT-B/16, Attack Protocol=PGD-10, Training time per epoch (s)=11122026.01 | 28.13 | 35.94 | |
| CLIPBackbone=ViT-B/32, Proxy CLIP=ViT-B/16, Attack Protocol=PGD-10, Training time per epoch (s)=02026.01 | 18.75 | 39.9 | |
| UCATFine-tuning dataset=TinyImageNet, Adversarial training=2-step PGD, Epsilon=1/255, Evaluation protocol=Zero-shot2025.12 | 18.14 | 31.92 | |
| FT-StandardBackbone=ViT-B/32, Proxy CLIP=ViT-B/16, Attack Protocol=PGD-10, Training time per epoch (s)=4062026.01 | 17.57 | 41.79 | |
| TRADESFine-tuning dataset=TinyImageNet, Adversarial training=2-step PGD, Epsilon=1/255, Evaluation protocol=Zero-shot2025.12 | 15.75 | 20.59 | |
| DKLFine-tuning dataset=TinyImageNet, Adversarial training=2-step PGD, Epsilon=1/255, Evaluation protocol=Zero-shot2025.12 | 15.69 | 21.17 | |
| ACATFine-tuning dataset=TinyImageNet, Adversarial training=2-step PGD, Epsilon=1/255, Evaluation protocol=Zero-shot2025.12 | 10.96 | 17.29 | |
| CLIPFine-tuning dataset=TinyImageNet, Adversarial training=2-step PGD, Epsilon=1/255, Evaluation protocol=Zero-shot2025.12 | 0.11 | 40.69 |