Image Classification on ImageNet (Robustness Metrics)
76.3Top-1 Accuracy (Clean)ResNet-50
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| ResNet-50Training Protocol=regular train2019.04 | 76.3 | — | — | |
| LR-Net-26Training Protocol=regular train2019.04 | 75.7 | — | — | |
| ResNet-26Training Protocol=regular train2019.04 | 72.8 | — | — | |
| LR-Net-26Training Protocol=adversarial train2019.04 | 52.1 | 44.2 | 26.8 | |
| ResNet-50Training Protocol=adversarial train2019.04 | 52 | 43 | 22.5 | |
| ResNet-26Training Protocol=adversarial train2019.04 | 44.9 | 37.9 | 14.4 |