Image Classification on CIFAR-10 (Metrics Nat, Union)
95.2Accuracy (Nat)MAX + RoME
Evaluation results
| Method | Links | ||
|---|---|---|---|
| MAX + RoMEArchitecture=WRN-94-16, Pre-training=l_inf-adversarial training2026.07 | 95.2 | 57.5 | |
| RAMP + RoMEArchitecture=WRN-94-16, Pre-training=l_inf-adversarial training2026.07 | 93.6 | 58.7 | |
| E-AT + RoMEArchitecture=WRN-94-16, Pre-training=l_inf-adversarial training2026.07 | 93.3 | 48.4 | |
| E-ATArchitecture=WRN-94-16, Pre-training=l_inf-adversarial training2026.07 | 92.8 | 44.7 | |
| E-AT + RoMEArchitecture=WRN-28-10, Pre-training=l_inf-adversarial training2026.07 | 91.3 | 51.8 | |
| RAMP + RoMEArchitecture=WRN-28-10, Pre-training=l_inf-adversarial training2026.07 | 91.3 | 54.2 | |
| MAXArchitecture=WRN-94-16, Pre-training=l_inf-adversarial training2026.07 | 91.2 | 51.4 | |
| RAMPArchitecture=WRN-94-16, Pre-training=l_inf-adversarial training2026.07 | 90.7 | 55.8 | |
| MAX + RoMEArchitecture=WRN-28-10, Pre-training=l_inf-adversarial training2026.07 | 90.5 | 53.7 | |
| E-ATArchitecture=WRN-28-10, Pre-training=l_inf-adversarial training2026.07 | 89.4 | 50.3 | |
| RAMPArchitecture=WRN-28-10, Pre-training=l_inf-adversarial training2026.07 | 89.2 | 52.3 | |
| MAXArchitecture=WRN-28-10, Pre-training=l_inf-adversarial training2026.07 | 88.7 | 51.4 | |
| E-AT + RoMEArchitecture=ViT-B, Pre-training=l_inf-adversarial training2026.07 | 84.7 | 43.5 | |
| E-ATArchitecture=ViT-B, Pre-training=l_inf-adversarial training2026.07 | 83.6 | 41.7 | |
| MAX + RoMEArchitecture=ViT-B, Pre-training=l_inf-adversarial training2026.07 | 83.6 | 44.4 | |
| RAMPArchitecture=ViT-B, Pre-training=l_inf-adversarial training2026.07 | 83.1 | 43.7 | |
| MAXArchitecture=ViT-B, Pre-training=l_inf-adversarial training2026.07 | 82.9 | 43.1 | |
| RAMP + RoMEArchitecture=ViT-B, Pre-training=l_inf-adversarial training2026.07 | 82.7 | 46.5 |