| ResNet50 (baseline), BatchNorm Adaptation, 8 samples | 65.0 | Improving robustness against common corruptions by covariate shift adaptation | |
| ResNet50 (baseline), BatchNorm Adaptation, full adaptation | 62.2 | Improving robustness against common corruptions by covariate shift adaptation | |
| ResNet50 + ENT | 51.6 | If your data distribution shifts, use self-learning | |
| ResNet50 + RPL | 50.5 | If your data distribution shifts, use self-learning | |
| ResNet50+DeepAug+AugMix, BatchNorm Adaptation, 8 samples | 48.4 | Improving robustness against common corruptions by covariate shift adaptation | |
| ResNet50+DeepAug+AugMix, BatchNorm Adaptation, full adaptation | 45.4 | Improving robustness against common corruptions by covariate shift adaptation | |
| ResNeXt101 32x8d + ENT | 44.3 | If your data distribution shifts, use self-learning | |
| ResNeXt101 32x8d + RPL | 43.2 | If your data distribution shifts, use self-learning | |
| ResNeXt101 32x8d + IG-3.5B + RPL | 40.9 | If your data distribution shifts, use self-learning | |
| ResNeXt101 32x8d + IG-3.5B + ENT | 40.8 | If your data distribution shifts, use self-learning | |
| ResNeXt101+DeepAug+AugMix, BatchNorm Adaptation, 8 samples | 40.7 | Improving robustness against common corruptions by covariate shift adaptation | |
| ResNeXt101+DeepAug+AugMix, BatchNorm Adaptation, full adaptation | 38.0 | Improving robustness against common corruptions by covariate shift adaptation | |
| ResNeXt101 32x8d + DeepAug + Augmix + ENT | 35.5 | If your data distribution shifts, use self-learning | |
| ResNeXt101 32x8d + DeepAug + Augmix + RPL | 34.8 | If your data distribution shifts, use self-learning | |
| EfficientNet-L2+ENT | 23.0 | If your data distribution shifts, use self-learning | |
| EfficientNet-L2+RPL | 22.0 | If your data distribution shifts, use self-learning | |