Robust Image Classification on ImageNet-1K (Zero-shot)
24.53Robust AccuracySAFT-L
Evaluation results
| Method | Links | ||
|---|---|---|---|
| SAFT-LAttack Protocol=C&W (ϵ = 1/255), Evaluation Setting=zero-shot2026.02 | 24.53 | 37.45 | |
| SAFT-LAttack Protocol=AutoAttack (ϵ = 1/255), Evaluation Setting=zero-shot2026.02 | 20.22 | 30.06 | |
| TGA-ZSRAttack Protocol=C&W (ϵ = 1/255), Evaluation Setting=zero-shot2026.02 | 19.89 | 29.94 | |
| TGA-ZSRAttack Protocol=AutoAttack (ϵ = 1/255), Evaluation Setting=zero-shot2026.02 | 19.2 | 28.89 | |
| PMG-AFTAttack Protocol=AutoAttack (ϵ = 1/255), Evaluation Setting=zero-shot2026.02 | 19.17 | 26.69 | |
| TeCoAAttack Protocol=AutoAttack (ϵ = 1/255), Evaluation Setting=zero-shot2026.02 | 16.21 | 23.64 | |
| TeCoAAttack Protocol=C&W (ϵ = 1/255), Evaluation Setting=zero-shot2026.02 | 14.24 | 23.93 | |
| PMG-AFTAttack Protocol=C&W (ϵ = 1/255), Evaluation Setting=zero-shot2026.02 | 12.92 | 25.83 |