Image Classification on ImageNet Linf epsilon=2/255 (test)
64.45Accuracy (Clean)Free-AT
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Free-ATBackbone=ResNet-50, Train ε=2/2552021.11 | 64.45 | 43.52 | 43.39 | 43.4 | |
| MixACM + RABackbone=ResNet-50, Train ε=02021.11 | 62.05 | 48.54 | 48.45 | 48.44 | |
| Fast-ATBackbone=ResNet-50, Train ε=2/2552021.11 | 60.9 | — | 43.46 | — | |
| MixACMBackbone=ResNet-50, Train ε=02021.11 | 59.64 | 46.03 | 45.96 | 45.92 |