Adversarial Zero-shot Image Classification on ImageNet (val)
61.7Adversarial AccuracyTAME
Evaluation Results
| Method | Links | |
|---|---|---|
| TAMEAttack Method=PGD, Backbone=ViT-B/16, Input Configuration=V-L Independent2026.05 | 61.7 | |
| TAMEAttack Method=AutoAttack, Backbone=ViT-B/16, Input Configuration=V-L Independent2026.05 | 60.8 | |
| TAMEAttack Method=CW, Backbone=ViT-B/16, Input Configuration=V-L Independent2026.05 | 60.7 | |
| TAMEAttack Method=DI, Backbone=ViT-B/16, Input Configuration=V-L Independent2026.05 | 59.4 | |
| R-TPTAttack Method=PGD, Backbone=ViT-B/162026.05 | 54.8 | |
| TAPTAttack Method=PGD, Backbone=ViT-B/16, Input Configuration=V-L Independent2026.05 | 50 | |
| MMoPAttack Method=PGD, Backbone=ViT-B/16, Input Configuration=V-L Independent2026.05 | 28 | |
| NAP-TuningAttack Method=PGD, Backbone=ViT-B/162026.05 | 27.2 | |
| APT-VLIAttack Method=PGD, Backbone=ViT-B/162026.05 | 24.3 | |
| APT-VLJAttack Method=PGD, Backbone=ViT-B/162026.05 | 23.9 | |
| FAPAttack Method=PGD, Backbone=ViT-B/162026.05 | 23.2 | |
| APT-VAttack Method=PGD, Backbone=ViT-B/162026.05 | 19.4 | |
| VanillaAttack Method=DI, Backbone=ViT-B/162026.05 | 6.1 | |
| APT-TAttack Method=PGD, Backbone=ViT-B/162026.05 | 1.5 | |
| VanillaAttack Method=PGD, Backbone=ViT-B/162026.05 | 1.4 | |
| VanillaAttack Method=CW, Backbone=ViT-B/162026.05 | 0.2 | |
| VanillaAttack Method=AutoAttack, Backbone=ViT-B/162026.05 | 0 |