mlsanityData to 7 Oct 2026

Revisiting Batch Normalization for Improving Corruption Robustness

arXiv 2010.03630 ↗First posted 7 Oct 2020cs.CVcited by 55

Not on the map. Why?

Abstract (arXiv)

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%.

Key facts. Cited by 55 papers in the map (data through 7 Oct 2026). First posted on arXiv 7 Oct 2020.

Builds on

No reference list yet.

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2021: 12; 2022: 14; 2023: 11; 2024: 10; 2025: 4; 2026: 4.

55 papers.

Sort:

By area:

  1. 2023A Comprehensive Survey on Test-Time Adaptation under Distribution Shiftscited by 353
  2. 2022Uncertainty Modeling for Out-of-Distribution Generalizationcited by 122
  3. 2021Test-Time Adaptation to Distribution Shift by Confidence Maximization and Input Transformationcited by 67
  4. 2021Rethinking "Batch" in BatchNormcited by 44
  5. 2022Partial and Asymmetric Contrastive Learning for Out-of-Distribution Detection in Long-Tailed Recognitioncited by 44
  6. 2021PRIME: A few primitives can boost robustness to common corruptionscited by 40
  7. 2023Robustness of SAM: Segment Anything Under Corruptions and Beyondcited by 30
  8. 2022Removing Batch Normalization Boosts Adversarial Trainingcited by 24
  9. 2023A Survey on the Robustness of Computer Vision Models against Common Corruptionscited by 20
  10. 2021Universal Adversarial Perturbations Through the Lens of Deep Steganography: Towards A Fourier Perspectivecited by 18
  11. 2023EAT: Towards Long-Tailed Out-of-Distribution Detectioncited by 16
  12. 2021Universal Adversarial Training with Class-Wise Perturbationscited by 13
38 more
  1. 2022Improving Robustness by Enhancing Weak Subnetscited by 12
  2. 2022Understanding the Failure of Batch Normalization for Transformers in NLPcited by 12
  3. 2023HybridAugment++: Unified Frequency Spectra Perturbations for Model Robustnesscited by 12
  4. 2024Benchmarking the Robustness of Temporal Action Detection Models Against Temporal Corruptionscited by 12
  5. 2021Anti-aliasing Deep Image Classifiers using Novel Depth Adaptive Blurring and Activation Functioncited by 11
  6. 2023StyLess: Boosting the Transferability of Adversarial Examplescited by 11
  7. 2024Test-time Adaptation for Regression by Subspace Alignmentcited by 9
  8. 2022Test-time Batch Normalizationcited by 8
  9. 2023Understanding the robustness difference between stochastic gradient descent and adaptive gradient methodscited by 8
  10. 2024Domain-Specific Block Selection and Paired-View Pseudo-Labeling for Online Test-Time Adaptationcited by 8
  11. 2022Delving into the Estimation Shift of Batch Normalization in a Networkcited by 7
  12. 2023REALM: Robust Entropy Adaptive Loss Minimization for Improved Single-Sample Test-Time Adaptationcited by 7
  13. 2024Un-Mixing Test-Time Normalization Statistics: Combatting Label Temporal Correlationcited by 7
  14. 2024Spectrum Extraction and Clipping for Implicitly Linear Layerscited by 7
  15. 2024Improving robustness to corruptions with multiplicative weight perturbationscited by 7
  16. 2022How Robust are Discriminatively Trained Zero-Shot Learning Models?cited by 6
  17. 2022Robustness to corruption in pre-trained Bayesian neural networkscited by 5
  18. 2022Benchmarking performance of object detection under image distortions in an uncontrolled environmentcited by 5
  19. 2021Maximum and Leaky Maximum Propagationcited by 4
  20. 2022Understanding the Robustness of Multi-Exit Models under Common Corruptionscited by 4
  21. 2022Learning Less Generalizable Patterns with an Asymmetrically Trained Double Classifier for Better Test-Time Adaptationcited by 3
  22. 2024Test-time Similarity Modification for Person Re-identification toward Temporal Distribution Shiftcited by 3
  23. 2025Robo-SGG: Exploiting Layout-Oriented Normalization and Restitution Can Improve Robust Scene Graph Generationcited by 3
  24. 2021Towards Bridging the gap between Empirical and Certified Robustness against Adversarial Examplescited by 1
  25. 2021Distribution Mismatch Correction for Improved Robustness in Deep Neural Networkscited by 1
  26. 2021UniFed: A Unified Framework for Federated Learning on Non-IID Image Featurescited by 1
  27. 2023On Sensitivity and Robustness of Normalization Schemes to Input Distribution Shifts in Automatic MR Image Diagnosiscited by 1
  28. 2023Backdoor Mitigation by Correcting the Distribution of Neural Activationscited by 1
  29. 2024LOTOS: Layer-wise Orthogonalization for Training Robust Ensemblescited by 1
  30. 2024Improving Batch Normalization with TTA for Robust Object Detection in Self-Drivingcited by 1
  31. 2025LeanTTA: A Backpropagation-Free and Stateless Approach to Quantized Test-Time Adaptation on Edge Devicescited by 1
  32. 2025EmbodiTTA: Resource-Efficient Test-Time Adaptation for Embodied Visual Systemscited by 1
  33. 2026Batch Normalization Amplifies Memorization and Privacy Riskscited by 1
  34. 2021Built-in Elastic Transformations for Improved Robustnesscited by 0
  35. 2021Wiggling Weights to Improve the Robustness of Classifierscited by 0
  36. 2022Counterbalancing Teacher: Regularizing Batch Normalized Models for Robustnesscited by 0
  37. 2022Adaptive Batch Normalization for Training Data with Heterogeneous Featurescited by 0
  38. 2023Dynamic Batch Norm Statistics Update for Natural Robustnesscited by 0

50 most cited of 55 All 55 (JSON)