Robustness Accuracy Evaluation on Tiny-ImageNet (test)
65.28Standard AccuracyWang et al. (2023) + S2O
Evaluation results
| Method | Links | ||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Wang et al. (2023) + S2OModel=WRN-28-10, DDPM Data=1M2026.03 | 65.28 | — | — | — | — | — | — | — | — | — | — | 31.6 | — | — | |
| Wang et al. (2023)Model=WRN-28-10, DDPM Data=1M2026.03 | 65.19 | — | — | — | — | — | — | — | — | — | — | 31.3 | — | — | |
| Cui et al. (2021)Model=WRN-28-10, DDPM Data=1M2026.03 | 60.95 | — | — | — | — | — | — | — | — | — | — | 26.66 | — | — | |
| SAAD-CStudent Model=PreActResNet-18, Teacher Model=WRN-28-10 (Wang2023Better)2025.12 | 57.33 | — | — | — | — | — | — | — | 33.06 | 29.62 | 24.16 | 22.69 | — | — | |
| AdaADStudent Model=PreActResNet-18, Teacher Model=WRN-28-10 (Wang2023Better)2025.12 | 57.26 | — | — | — | — | — | — | — | 31.81 | 28.8 | 23.64 | 22.11 | — | — | |
| SAADStudent Model=PreActResNet-18, Teacher Model=WRN-28-10 (Wang2023Better)2025.12 | 57.16 | — | — | — | — | — | — | — | 33.26 | 29.95 | 24.87 | 23.42 | — | — | |
| IGDMStudent Model=PreActResNet-18, Teacher Model=WRN-28-10 (Wang2023Better)2025.12 | 57.15 | — | — | — | — | — | — | — | 31.98 | 29.02 | 23.94 | 22.52 | — | — | |
| ARDStudent Model=PreActResNet-18, Teacher Model=WRN-28-10 (Wang2023Better)2025.12 | 55.69 | — | — | — | — | — | — | — | 30.13 | 26.62 | 22.09 | 19.9 | — | — | |
| AKDStudent Model=PreActResNet-18, Teacher Model=WRN-28-10 (Wang2023Better)2025.12 | 54.48 | — | — | — | — | — | — | — | 27.62 | 23.59 | 19.56 | 17.73 | — | — | |
| IADStudent Model=PreActResNet-18, Teacher Model=WRN-28-10 (Wang2023Better)2025.12 | 53.75 | — | — | — | — | — | — | — | 29.85 | 26.95 | 22.4 | 20.56 | — | — | |
| RSLADStudent Model=PreActResNet-18, Teacher Model=WRN-28-10 (Wang2023Better)2025.12 | 53.18 | — | — | — | — | — | — | — | 30.14 | 27.69 | 22.86 | 21.42 | — | — | |
| RS-FGSMnoise magnitude=8/255, backbone=PreActResNet-18, threat model=L∞2024.04 | 52.28 | — | — | — | — | — | — | 0 | — | — | — | — | — | — | |
| TRADESBackbone=WideResNet-28-102026.04 | 51.14 | — | — | — | — | — | — | — | — | — | 22.75 | 20.02 | 24.84 | 24.58 | |
| Cons-ATBackbone=WideResNet-28-102026.04 | 50.12 | — | — | — | — | — | — | — | — | — | 22.17 | 20.64 | 25.05 | 24.42 | |
| RS-AAERnoise magnitude=8/255, backbone=PreActResNet-18, threat model=L∞2024.04 | 49.86 | — | — | — | — | — | — | 19.66 | — | — | — | — | — | — | |
| RAATBackbone=WideResNet-28-102026.04 | 49.32 | — | — | — | — | — | — | — | — | — | 22.28 | 21.3 | 25.41 | 24.65 | |
| N-FGSMnoise magnitude=8/255, backbone=PreActResNet-18, threat model=L∞2024.04 | 48.16 | — | — | — | — | — | — | 20.73 | — | — | — | — | — | — | |
| RAAT++Backbone=WideResNet-28-102026.04 | 47.96 | — | — | — | — | — | — | — | — | — | 23.78 | 22.12 | 26.54 | 26.01 | |
| N-AAERnoise magnitude=8/255, backbone=PreActResNet-18, threat model=L∞2024.04 | 47.93 | — | — | — | — | — | — | 20.92 | — | — | — | — | — | — | |
| PGD-ATBackbone=WideResNet-28-102026.04 | 47.79 | — | — | — | — | — | — | — | — | — | 21.48 | 20 | 23.97 | 23.59 | |
| RAATBackbone=ResNet-182026.04 | 46.77 | — | — | — | — | — | — | — | — | — | 19.97 | 17.88 | 22.63 | 22.28 | |
| TRADESBackbone=ResNet-182026.04 | 46.75 | — | — | — | — | — | — | — | — | — | 18.83 | 16.6 | 21.62 | 21.52 | |
| Cons-ATBackbone=ResNet-182026.04 | 46.54 | — | — | — | — | — | — | — | — | — | 19.88 | 17.6 | 22.58 | 21.7 | |
| PGD-2noise magnitude=8/255, backbone=PreActResNet-18, threat model=L∞2024.04 | 46.43 | — | — | — | — | — | — | 20.72 | — | — | — | — | — | — | |
| PGD-ATBackbone=ResNet-182026.04 | 46.32 | — | — | — | — | — | — | — | — | — | 18.79 | 17.07 | 21.75 | 21.52 | |
| PGD-ATStudent Model=PreActResNet-18, Teacher Model=WRN-28-10 (Wang2023Better)2025.12 | 45.71 | — | — | — | — | — | — | — | 15.75 | 11.67 | 11.82 | 10.91 | — | — | |
| MARTBackbone=WideResNet-28-102026.04 | 45.57 | — | — | — | — | — | — | — | — | — | 23.11 | 21.07 | 26.21 | 25.92 | |
| TRADESStudent Model=PreActResNet-18, Teacher Model=WRN-28-10 (Wang2023Better)2025.12 | 42.06 | — | — | — | — | — | — | — | 19.58 | 17.15 | 14.03 | 13.33 | — | — | |
| RAAT++Backbone=ResNet-182026.04 | 41.69 | — | — | — | — | — | — | — | — | — | 20.62 | 17.21 | 22.92 | 22.83 | |
| MARTBackbone=ResNet-182026.04 | 39.7 | — | — | — | — | — | — | — | — | — | 19.26 | 17.18 | 22.98 | 22.79 | |
| ATSetting=AT2021.10 | — | 17.2 | 15.4 | 1.8 | 47.72 | 47.62 | 0.1 | — | — | — | — | — | — | — | |
| KD-ATSetting=KD-AT, T=2, lambda=0.52021.10 | — | 17.86 | 17.18 | 0.68 | 47.73 | 48.28 | -0.55 | — | — | — | — | — | — | — | |
| KD-AT-AutoSetting=KD-AT-Auto, T=1.23*, lambda=0.85*2021.10 | — | 18.29 | 18.39 | -0.1 | 47.46 | 47.56 | -0.1 | — | — | — | — | — | — | — |