Improving robustness against common corruptions by covariate shift adaptation

Benchmark Model Rank Results
image-classification-on-objectnetResNet-50 + GroupNorm#60Top-1 Accuracy: 29.2Top-5 Accuracy: 50.2
image-classification-on-objectnetResNet-50 + RoHL#61Top-1 Accuracy: 29.2
image-classification-on-objectnetResNet-50 + FixUp#64Top-1 Accuracy: 28.5Top-5 Accuracy: 48.6
unsupervised-domain-adaptation-on-imagenet-cResNeXt101+DeepAug+AugMix, BatchNorm Adaptation, full adaptation#5mean Corruption Error (mCE): 38.0
unsupervised-domain-adaptation-on-imagenet-cResNeXt101+DeepAug+AugMix, BatchNorm Adaptation, 8 samples#6mean Corruption Error (mCE): 40.7
unsupervised-domain-adaptation-on-imagenet-cResNet50+DeepAug+AugMix, BatchNorm Adaptation, full adaptation#11mean Corruption Error (mCE): 45.4
unsupervised-domain-adaptation-on-imagenet-cResNet50+DeepAug+AugMix, BatchNorm Adaptation, 8 samples#12mean Corruption Error (mCE): 48.4
unsupervised-domain-adaptation-on-imagenet-cResNet50 (baseline), BatchNorm Adaptation, full adaptation#15mean Corruption Error (mCE): 62.2
unsupervised-domain-adaptation-on-imagenet-cResNet50 (baseline), BatchNorm Adaptation, 8 samples#16mean Corruption Error (mCE): 65.0
unsupervised-domain-adaptation-on-imagenet-rResNeXt101+DeepAug+AugMix, BatchNorm Adaptation,#4Top 1 Error: 44.0
unsupervised-domain-adaptation-on-imagenet-rResNet50+DeepAug+Augmix, BatchNorm adaptation#5Top 1 Error: 48.9
unsupervised-domain-adaptation-on-imagenet-rResNet50, BatchNorm adaptation#8Top 1 Error: 59.9