Adversarial Robust Image Classification on CIFAR-10 (test) (Clean and AA Metrics)
88.71Clean AccuracyDAJAT
Evaluation results
| Method | Links | ||
|---|---|---|---|
| DAJATBackbone=WRN-34-102022.10 | 88.71 | 57.81 | |
| SEATBackbone=WRN-34-102022.10 | 86.44 | 55.67 | |
| HATBackbone=WRN-34-102022.10 | 86.21 | 51.46 | |
| DAJATBackbone=ResNet-182022.10 | 85.71 | 52.5 | |
| HATBackbone=ResNet-182022.10 | 85.63 | 49.54 | |
| AWPBackbone=WRN-34-102022.10 | 85.36 | 56.17 | |
| UDR + TRADESBackbone=WRN-34-102022.10 | 84.93 | 54.45 | |
| SEAT+CutmixBackbone=WRN-34-102022.10 | 84.81 | 56.03 | |
| UDR + TRADESBackbone=ResNet-182022.10 | 84.4 | 49.9 | |
| TRADES + TEBackbone=ResNet-182022.10 | 83.86 | 49.77 | |
| SEATBackbone=ResNet-182022.10 | 83.7 | 51.3 | |
| AWPBackbone=ResNet-182022.10 | 81.99 | 51.45 | |
| SEAT+CutmixBackbone=ResNet-182022.10 | 81.53 | 49.1 |