Adversarial Robustness on CIFAR-100 (PGD, C&W, AA Metrics)
66.02Clean AccuracyHICAT
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| HICATBackbone=WRN28-10, Data Augmentation=Diffusion-generated samples2026.06 | 66.02 | — | — | — | — | 36.2 | 2.51 | |
| DHATBackbone=WRN28-10, Data Augmentation=Diffusion-generated samples2026.06 | 64.8 | — | — | — | — | 34.71 | 2.96 | |
| HICATBackbone=ResNet-18, Data Augmentation=Diffusion-generated samples2026.06 | 64.52 | — | — | — | — | 33.48 | 0.86 | |
| DHATBackbone=ResNet-18, Data Augmentation=Diffusion-generated samples2026.06 | 63.14 | — | — | — | — | 32.21 | 0.98 | |
| SGLRBackbone=WRN28-10, Data Augmentation=Diffusion-generated samples2026.06 | 62.39 | — | — | — | — | 30.15 | 17.05 | |
| HICATBackbone=WRN28-10, Attack perturbation budget (epsilon)=8/2552026.06 | 62.38 | 37.29 | 36.82 | 36.71 | 33.15 | 31.24 | 3.12 | |
| CFABackbone=WRN28-10, Data Augmentation=Diffusion-generated samples2026.06 | 61.56 | — | — | — | — | 29.61 | 9.13 | |
| SGLRBackbone=ResNet-18, Data Augmentation=Diffusion-generated samples2026.06 | 61.25 | — | — | — | — | 29.1 | 4.15 | |
| UIATBackbone=WRN28-10, Data Augmentation=Diffusion-generated samples2026.06 | 60.24 | — | — | — | — | 30.98 | 14.32 | |
| CFABackbone=ResNet-18, Data Augmentation=Diffusion-generated samples2026.06 | 60.13 | — | — | — | — | 28.85 | 3.01 | |
| UIATBackbone=ResNet-18, Data Augmentation=Diffusion-generated samples2026.06 | 59.92 | — | — | — | — | 28.48 | 4.47 | |
| AWPBackbone=WRN28-10, Data Augmentation=Diffusion-generated samples2026.06 | 59.71 | — | — | — | — | 31.83 | 7.13 | |
| FSRBackbone=WRN28-10, Data Augmentation=Diffusion-generated samples2026.06 | 59.03 | — | — | — | — | 30.44 | 7.4 | |
| FSRBackbone=ResNet-18, Data Augmentation=Diffusion-generated samples2026.06 | 58.1 | — | — | — | — | 28.94 | 2.47 | |
| AWPBackbone=ResNet-18, Data Augmentation=Diffusion-generated samples2026.06 | 57.55 | — | — | — | — | 29.33 | 2.37 | |
| MARTBackbone=WRN28-10, Data Augmentation=Diffusion-generated samples2026.06 | 55.87 | — | — | — | — | 30.2 | 8.86 | |
| MARTBackbone=ResNet-18, Data Augmentation=Diffusion-generated samples2026.06 | 54.73 | — | — | — | — | 27.7 | 2.88 | |
| MARTBackbone=WRN28-10, Attack perturbation budget (epsilon)=8/2552026.06 | 54.69 | 32.06 | 31.9 | 31.88 | 28.77 | 27.25 | 9.96 |