Image Classification on CIFAR-10 (test) (Transformation Robustness)
83.93Accuracy (Base)Clean
Evaluation results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| Clean2026.01 | 83.93 | 91.45 | 92.05 | 93.38 | 88.84 | 91.99 | 92.71 | 93.49 | 93.87 | |
| CUDA*Perturbation budget (epsilon)=8/2552026.01 | 65.47 | 87.45 | 89.49 | 88.54 | 80.86 | 80.82 | 85.8 | 81.34 | 89.05 | |
| TAP2026.01 | 35.9 | 19.11 | 21.18 | 21.47 | 19.38 | 13.56 | 15.09 | 11.64 | 20.3 | |
| SEP2026.01 | 28.43 | 8.94 | 19.68 | 23.7 | 9.61 | 9.8 | 9.74 | 12.02 | 10.48 | |
| NTGA2026.01 | 27.6 | 30.22 | 29.45 | 31.42 | 70.55 | 56.21 | 25.86 | 15.8 | 17.34 | |
| EM2026.01 | 21.43 | 24.82 | 28.67 | 29.77 | 89.15 | 57.64 | 23.47 | 29.39 | 48.26 | |
| REM2026.01 | 21.36 | 23.24 | 22.71 | 26.16 | 69.35 | 64.37 | 17.82 | 24.46 | 24.91 | |
| AR2026.01 | 16.89 | 17.57 | 11.31 | 13.4 | 38.38 | 14.25 | 10.57 | 14.05 | 15.87 | |
| SP2026.01 | 16.06 | 23.92 | 23.67 | 25.28 | 82.79 | 73.04 | 23.74 | 21.11 | 26.41 | |
| PIL2026.01 | 14.7 | 12.87 | 18.15 | 19.3 | 17.01 | 10.88 | 14.62 | 10.79 | 11.05 | |
| CUDAPerturbation budget=original larger budget2026.01 | 10.35 | 24.74 | 29.53 | 30.7 | 23.74 | 25.72 | 21.62 | 23.95 | 23.54 |