English

Revisiting Batch Normalization for Improving Corruption Robustness

Computer Vision and Pattern Recognition 2021-01-29 v4

Abstract

The performance of DNNs trained on clean images has been shown to decrease when the test images have common corruptions. In this work, we interpret corruption robustness as a domain shift and propose to rectify batch normalization (BN) statistics for improving model robustness. This is motivated by perceiving the shift from the clean domain to the corruption domain as a style shift that is represented by the BN statistics. We find that simply estimating and adapting the BN statistics on a few (32 for instance) representation samples, without retraining the model, improves the corruption robustness by a large margin on several benchmark datasets with a wide range of model architectures. For example, on ImageNet-C, statistics adaptation improves the top1 accuracy of ResNet50 from 39.2% to 48.7%. Moreover, we find that this technique can further improve state-of-the-art robust models from 58.1% to 63.3%.

Keywords

Cite

@article{arxiv.2010.03630,
  title  = {Revisiting Batch Normalization for Improving Corruption Robustness},
  author = {Philipp Benz and Chaoning Zhang and Adil Karjauv and In So Kweon},
  journal= {arXiv preprint arXiv:2010.03630},
  year   = {2021}
}

Comments

Accepted at WACV 2021

R2 v1 2026-06-23T19:08:47.498Z