Image Classification on ImageNet (val) (PGD Robustness Accuracy)
76.038Accuracy (Natural Images)Natural
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| NaturalBackbone=ResNet-50, epsilon=22019.04 | 76.038 | 0.166 | 0.052 | 0.036 | |
| Free Adversarial Trainingm=2, Backbone=ResNet-50, epsilon=22019.04 | 71.21 | 37.012 | 36.34 | 36.25 | |
| Free Adversarial Trainingm=4, Backbone=ResNet-50, epsilon=22019.04 | 64.446 | 43.522 | 43.392 | 43.404 | |
| Free Adversarial Trainingm=6, Backbone=ResNet-50, epsilon=22019.04 | 60.642 | 41.996 | 41.9 | 41.892 | |
| Free Adversarial Trainingm=8, Backbone=ResNet-50, epsilon=22019.04 | 58.116 | 40.044 | 40.008 | 39.996 |