Image Classification on ImageNet full (val)
78.18Top-1 AccuracySwin-L
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Swin-LDefense=Base2026.03 | 78.18 | — | — | |
| ConvNeXt-LDefense=Base2026.03 | 77.47 | — | — | |
| Swin-LDefense=+ ET32026.03 | 77.21 | — | — | |
| ConvNeXt-BDefense=Base2026.03 | 76.38 | — | — | |
| ConvNeXt-LDefense=+ ET32026.03 | 76.36 | — | — | |
| Swin-BDefense=Base2026.03 | 76.21 | — | — | |
| ConvNeXt-BDefense=+ ET32026.03 | 75.95 | — | — | |
| Swin-BDefense=+ ET32026.03 | 75.75 | — | — | |
| ResNet-50Defense=Base2026.03 | 64.02 | — | — | |
| ResNet-50Defense=+ ET32026.03 | 63.12 | — | — | |
| Pre-act ResNet-200Testing image size=224x224, Crop protocol=single-crop2017.09 | — | 21.7 | 5.8 | |
| ResNet-152Testing image size=224x224, Crop protocol=single-crop2017.09 | — | 22.2 | 6.2 | |
| ResNet-152 with ERER proportion=50%, Testing image size=224x224, Crop protocol=single-crop2017.09 | — | 21.6 | 5.8 | |
| ResNet-200Testing image size=224x224, Crop protocol=single-crop, Note=Results run by authors2017.09 | — | 21.8 | 6.1 | |
| ResNet-200 with ERER proportion=50%, Testing image size=224x224, Crop protocol=single-crop2017.09 | — | 21.4 | 5.8 |