Adversarial Robustness Classification on CIFAR-10 (EOT Metrics)
80.58Robust Accuracy (C&W + EOT)RAMRM-PureMaskDiT
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| RAMRM-PureMaskDiTArchitecture=WideResNet-70-16, Norm setting=l2, Epsilon=1, Gradient type=exact gradient2026.07 | 80.58 | 86.11 | 73.59 | 69.57 | |
| AMRM-PureMaskDiTArchitecture=WideResNet-70-16, Norm setting=l2, Epsilon=1, Gradient type=exact gradient2026.07 | 79.15 | 82.8 | 66.43 | 64.53 | |
| ADBMArchitecture=WideResNet-70-16, Norm setting=l2, Epsilon=1, Gradient type=exact gradient2026.07 | 78.4 | 84.3 | 66.8 | 66.3 | |
| DiffpureArchitecture=WideResNet-70-16, Norm setting=l2, Epsilon=1, Gradient type=exact gradient2026.07 | 74.8 | 78.4 | 63.9 | 60.8 | |
| RAMRM-PureMAEArchitecture=WideResNet-70-16, Norm setting=l2, Epsilon=1, Gradient type=exact gradient2026.07 | 72.2 | 72.29 | 59.11 | 60.72 | |
| AMRM-PureMAEArchitecture=WideResNet-70-16, Norm setting=l2, Epsilon=1, Gradient type=exact gradient2026.07 | 62.54 | 65.15 | 52.75 | 53.5 |