Adversarial Attack Detection (test)
99.64FGSM AccuracyA^4D
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| A^4DModel=Wide ResNet, Setting=Zero-shot2026.06 | 99.64 | 99.67 | 99.75 | 99.44 | 78.03 | 77.55 | 93.61 | 92.53 | |
| WaveletModel=Wide ResNet, Setting=Zero-shot2026.06 | 99.64 | 97.43 | 97.68 | 82.88 | 50.48 | 50.11 | 70.28 | 78.36 | |
| WaveletModel=ResNet34, Setting=Zero-shot2026.06 | 99.58 | 97.42 | 97.63 | 82.49 | 50.46 | 50.02 | 70.46 | 78.29 | |
| A^4DModel=Deit-Small, Setting=Zero-shot2026.06 | 99.24 | 98.16 | 98.39 | 97.76 | 77.31 | 78.62 | 97.84 | 92.47 | |
| MahalanobisModel=Convnext tiny, Setting=Zero-shot2026.06 | 96.46 | 97.28 | — | 97.65 | 74.64 | 75.13 | 91.98 | 88.85 | |
| Mahalanobis*Model=Convnext tiny, Setting=Zero-shot, Variant=Class agnostic2026.06 | 96.46 | 98.29 | — | 98.27 | 59.83 | 57.48 | 90.25 | 83.43 | |
| WaveletModel=Deit-Small, Setting=Zero-shot2026.06 | 95.99 | 95.05 | 93.69 | 90.36 | 50.92 | 50.62 | 56.88 | 76.22 | |
| A^4DModel=ResNet34, Setting=Zero-shot2026.06 | 94.7 | 98.01 | 96.79 | 96.27 | 86.27 | 85.64 | 95.57 | 93.32 | |
| WaveletModel=Convnext tiny, Setting=Zero-shot2026.06 | 94.07 | 87.33 | 87.68 | 69.88 | 50.39 | 50.24 | 70.44 | 72.86 | |
| A^4DModel=Convnext tiny, Setting=Zero-shot2026.06 | 93.8 | 95.38 | 97.87 | 83.18 | 62.83 | 63.47 | 98.24 | 84.97 | |
| MahalanobisModel=ResNet34, Setting=Zero-shot2026.06 | 91.94 | 91.52 | 85.05 | 77.27 | 70.99 | 68.58 | 67.22 | 78.94 | |
| Mahalanobis*Model=ResNet34, Setting=Zero-shot, Variant=Class agnostic2026.06 | 91.94 | 13.14 | 29.33 | 18.59 | 38.57 | 38.24 | 42.61 | 38.92 | |
| MADModel=Wide ResNet, Setting=Zero-shot2026.06 | 85.8 | 79.08 | 82.36 | 62.94 | 50.02 | 49.98 | 81.45 | 70.23 | |
| MADModel=ResNet34, Setting=Zero-shot2026.06 | 85.55 | 78.85 | 82.2 | 63.34 | 50.17 | 50 | 82.94 | 70.43 | |
| MADModel=Convnext tiny, Setting=Zero-shot2026.06 | 77.73 | 69.64 | 72.29 | 57.12 | 50.1 | 50.01 | 76.65 | 64.79 | |
| MADModel=Deit-Small, Setting=Zero-shot2026.06 | 73.46 | 71.07 | 85.95 | 83.46 | 50.68 | 50.46 | 63.52 | 68.37 | |
| MahalanobisModel=Wide ResNet, Setting=Zero-shot2026.06 | 67.1 | 99.91 | 99.24 | 99.76 | 55.26 | 55.44 | 62.08 | 76.97 | |
| Mahalanobis*Model=Wide ResNet, Setting=Zero-shot, Variant=Class agnostic2026.06 | 67.1 | 82.91 | 81.6 | 78.64 | 54.65 | 55.1 | 50.02 | 67.15 | |
| MahalanobisModel=Deit-Small, Setting=Zero-shot2026.06 | 61.45 | 61.94 | — | 62.53 | 63.82 | 63.55 | 69.08 | 63.72 | |
| Mahalanobis*Model=Deit-Small, Setting=Zero-shot, Variant=Class agnostic2026.06 | 61.45 | 60.59 | — | 60.51 | 47.06 | 48.27 | 51.92 | 54.96 |