Image Classification on CIFAR-100 (Clean, GAMA, and AutoAttack Robustness)
70.35Clean AccuracyDAJAT (Base, 3*AA)
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| DAJAT (Base, 3*AA)No. Steps=2+6, Backbone=WideResNet-342022.10 | 70.35 | 31.15 | 30.89 | |
| DAJAT (Base, 2*AA)No. Steps=2+4, Backbone=WideResNet-342022.10 | 68.74 | 31.58 | 31.3 | |
| DAJAT (Base, AA)No. Steps=2+2, Backbone=WideResNet-342022.10 | 67.82 | 31.65 | 31.26 | |
| DAJAT (Base, 3*AA)No. Steps=2+6, Backbone=ResNet-182022.10 | 66.96 | 27.9 | 27.62 | |
| DAJAT (Base, 2*AA)No. Steps=2+4, Backbone=ResNet-182022.10 | 66.84 | 27.61 | 27.32 | |
| DAJAT (Base, AA)No. Steps=2+2, Backbone=ResNet-182022.10 | 65.75 | 27.58 | 27.21 | |
| ACAT (Base, 2step)No. Steps=2, Backbone=WideResNet-342022.10 | 65.75 | 30.61 | 30.23 | |
| TRADES-AWP-WANo. Steps=10, Backbone=WideResNet-342022.10 | 62.73 | 29.92 | 29.59 | |
| TRADES-AWPNo. Steps=10, Backbone=WideResNet-342022.10 | 62.41 | 29.7 | 29.54 | |
| ACAT (Base, 2step)No. Steps=2, Backbone=ResNet-182022.10 | 62.05 | 26.35 | 26.1 | |
| TRADES-AWP-WANo. Steps=10, Backbone=ResNet-182022.10 | 59.88 | 25.81 | 25.52 | |
| TRADES-AWPNo. Steps=10, Backbone=ResNet-182022.10 | 58.81 | 25.51 | 25.3 |