Adversarial Attack Perturbation Analysis on ImageNet
72.04ParsimonPyTorch ResNet-50
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| PyTorch ResNet-50source=PyTorch, Architecture=ResNet-50, Iterations=10002025.11 | 72.04 | 84.64 | 80.8 | 55.8 | 87.28 | 2.64 | |
| Salman ResNet-18source=Salman, Architecture=ResNet-18, Iterations=10002025.11 | 58.6 | 58.72 | 54.92 | 54.26 | 61.98 | 3.26 | |
| Wong ResNet-50source=Wong, Architecture=ResNet-50, Iterations=10002025.11 | 56.62 | 56.62 | 52.46 | 50.34 | 59.96 | 3.34 | |
| Engstrom ResNet-50source=Engstrom, Architecture=ResNet-50, Iterations=10002025.11 | 48.92 | 48.16 | 45.1 | 44.12 | 51.84 | 2.92 | |
| Salman ResNet-50source=Salman, Architecture=ResNet-50, Iterations=10002025.11 | 46.96 | 46.7 | 44.06 | 43.08 | 50.16 | 3.2 | |
| Salman WideResNet-50-2source=Salman, Architecture=WideResNet-50-2, Iterations=10002025.11 | 42.94 | 42.22 | 39.66 | 38.32 | 44.86 | 1.92 | |
| Debenedetti XCiT-S12source=Debenedetti, Architecture=XCiT-S12, Iterations=10002025.11 | 37.44 | 36.48 | 33.74 | 32.96 | 39.66 | 2.22 | |
| Debenedetti XCiT-M12source=Debenedetti, Architecture=XCiT-M12, Iterations=10002025.11 | 36.04 | 35.1 | 32.7 | 31.86 | 37.64 | 1.6 | |
| Singh ViT-S+ConvStemsource=Singh, Architecture=ViT-S+ConvStem, Iterations=10002025.11 | 35.5 | 35.3 | 32.68 | 32.44 | 37.58 | 2.08 | |
| Debenedetti XCiT-L12source=Debenedetti, Architecture=XCiT-L12, Iterations=10002025.11 | 35.32 | 34.64 | 32.38 | 31.52 | 37.02 | 1.7 | |
| Singh ConvNeXt-T+ConvStemsource=Singh, Architecture=ConvNeXt-T+ConvStem, Iterations=10002025.11 | 34.88 | 34.5 | 32.24 | 31.58 | 37.12 | 2.24 | |
| Singh ConvNeXt-S+ConvStemsource=Singh, Architecture=ConvNeXt-S+ConvStem, Iterations=10002025.11 | 33.36 | 32.8 | 30.94 | 30.62 | 35.28 | 1.92 | |
| Singh ViT-B+ConvStemsource=Singh, Architecture=ViT-B+ConvStem, Iterations=10002025.11 | 31.02 | 30.38 | 28.76 | 28.2 | 32.56 | 1.54 | |
| Liu Swin-Bsource=Liu, Architecture=Swin-B, Iterations=10002025.11 | 30.88 | 30.44 | 28.64 | 28.08 | 32.6 | 1.72 | |
| Singh ConvNeXt-B+ConvStemsource=Singh, Architecture=ConvNeXt-B+ConvStem, Iterations=10002025.11 | 30.78 | 30.14 | 28.62 | 28.24 | 32.44 | 1.66 | |
| Liu ConvNeXt-Bsource=Liu, Architecture=ConvNeXt-B, Iterations=10002025.11 | 30.32 | 29.76 | 28.22 | 27.62 | 31.94 | 1.62 | |
| Singh ConvNeXt-L+ConvStemsource=Singh, Architecture=ConvNeXt-L+ConvStem, Iterations=10002025.11 | 29.24 | 28.8 | 27.3 | 26.42 | 30.64 | 1.4 | |
| Liu ConvNeXt-Lsource=Liu, Architecture=ConvNeXt-L, Iterations=10002025.11 | 28.92 | 28.34 | 26.86 | 26.4 | 30.2 | 1.28 | |
| Liu Swin-Lsource=Liu, Architecture=Swin-L, Iterations=10002025.11 | 28.18 | 27.44 | 25.86 | 25.32 | 29.9 | 1.72 |