mlsanityData to 7 Oct 2026

Improving robustness against common corruptions by covariate shift adaptation

arXiv 2006.16971 ↗First posted 30 Jun 2020cs.LG cs.CV stat.MLcited by 427

Not on the map. Why?

Cited by 25 in Multimodal LLMs / vision-language models, but placed by its own abstract, which didn’t fit.

Abstract (arXiv)

Today's state-of-the-art machine vision models are vulnerable to image corruptions like blurring or compression artefacts, limiting their performance in many real-world applications. We here argue that popular benchmarks to measure model robustness against common corruptions (like ImageNet-C) underestimate model robustness in many (but not all) application scenarios. The key insight is that in many scenarios, multiple unlabeled examples of the corruptions are available and can be used for unsupervised online adaptation. Replacing the activation statistics estimated by batch normalization on the training set with the statistics of the corrupted images consistently improves the robustness across 25 different popular computer vision models. Using the corrected statistics, ResNet-50 reaches 62.2% mCE on ImageNet-C compared to 76.7% without adaptation. With the more robust DeepAugment+AugMix model, we improve the state of the art achieved by a ResNet50 model up to date from 53.6% mCE to 45.4% mCE. Even adapting to a single sample improves robustness for the ResNet-50 and AugMix models, and 32 samples are sufficient to improve the current state of the art for a ResNet-50 architecture. We argue that results with adapted statistics should be included whenever reporting scores in corruption benchmarks and other out-of-distribution generalization settings.

Key facts. Cited by 427 papers in the map (data through 7 Oct 2026). First posted on arXiv 30 Jun 2020.

Builds on

No reference list yet.

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2021: 40; 2022: 67; 2023: 93; 2024: 95; 2025: 77; 2026: 55.

427 papers.

Sort:

By area:

  1. 2022Continual Test-Time Domain Adaptationcited by 520
  2. 2022Efficient Test-Time Model Adaptation without Forgettingcited by 467
  3. 2023Towards Stable Test-Time Adaptation in Dynamic Wild Worldcited by 409
  4. 2022Test-Time Prompt Tuning for Zero-Shot Generalization in Vision-Language Modelscited by 379
  5. 2021MEMO: Test Time Robustness via Adaptation and Augmentationcited by 376
  6. 2023A Comprehensive Survey on Test-Time Adaptation under Distribution Shiftscited by 353
  7. 2022Parameter-free Online Test-time Adaptationcited by 184
  8. 2021A Fine-Grained Analysis on Distribution Shiftcited by 171
  9. 2022NOTE: Robust Continual Test-time Adaptation Against Temporal Correlationcited by 150
  10. 2023Diverse Data Augmentation with Diffusions for Effective Test-time Prompt Tuningcited by 150
  11. 2024Efficient Test-Time Adaptation of Vision-Language Modelscited by 138
  12. 2021The Norm Must Go On: Dynamic Unsupervised Domain Adaptation by Normalizationcited by 131
38 more
  1. 2023EcoTTA: Memory-Efficient Continual Test-time Adaptation via Self-distilled Regularizationcited by 123
  2. 2022Robust Mean Teacher for Continual and Gradual Test-Time Adaptationcited by 102
  3. 2023TTN: A Domain-Shift Aware Batch Normalization in Test-Time Adaptationcited by 101
  4. 2023Align Your Prompts: Test-Time Prompting with Distribution Alignment for Zero-Shot Generalizationcited by 100
  5. 2021On Interaction Between Augmentations and Corruptions in Natural Corruption Robustnesscited by 98
  6. 2022Test-Time Adaptation via Conjugate Pseudo-labelscited by 98
  7. 2022Decorate the Newcomers: Visual Domain Prompt for Continual Test Time Adaptationcited by 98
  8. 2025Maximizing Confidence Alone Improves Reasoningcited by 97
  9. 2023Batch Calibration: Rethinking Calibration for In-Context Learning and Prompt Engineeringcited by 91
  10. 2022Benchmarking Robustness of 3D Point Cloud Recognition Against Common Corruptionscited by 78
  11. 2024Test-Time Model Adaptation with Only Forward Passescited by 74
  12. 2023DELTA: degradation-free fully test-time adaptationcited by 72
  13. 2021On the effectiveness of adversarial training against common corruptionscited by 71
  14. 2022Back to the Source: Diffusion-Driven Test-Time Adaptationcited by 71
  15. 2021Test-Time Adaptation to Distribution Shift by Confidence Maximization and Input Transformationcited by 67
  16. 2021Improved OOD Generalization via Adversarial Training and Pre-trainingcited by 66
  17. 2023Universal Test-time Adaptation through Weight Ensembling, Diversity Weighting, and Prior Correctioncited by 65
  18. 2022Test-Time Adaptation via Self-Training with Nearest Neighbor Informationcited by 64
  19. 2022Challenges and Opportunities in Offline Reinforcement Learning from Visual Observationscited by 63
  20. 2023On Pitfalls of Test-Time Adaptationcited by 63
  21. 2023SoTTA: Robust Test-Time Adaptation on Noisy Data Streamscited by 63
  22. 2023ViDA: Homeostatic Visual Domain Adapter for Continual Test Time Adaptationcited by 58
  23. 2021SITA: Single Image Test-time Adaptationcited by 57
  24. 2022An Efficient Domain-Incremental Learning Approach to Drive in All Weather Conditionscited by 54
  25. 2021Source-Free Adaptation to Measurement Shift via Bottom-Up Feature Restorationcited by 50
  26. 2021Test-time Batch Statistics Calibration for Covariate Shiftcited by 50
  27. 2023RDumb: A simple approach that questions our progress in continual test-time adaptationcited by 50
  28. 2023AdaNPC: Exploring Non-Parametric Classifier for Test-Time Adaptationcited by 48
  29. 2025Test-Time Learning for Large Language Modelscited by 46
  30. 2023Improved Test-Time Adaptation for Domain Generalizationcited by 45
  31. 2021Rethinking "Batch" in BatchNormcited by 44
  32. 2022Simple Cues Lead to a Strong Multi-Object Trackercited by 44
  33. 2022A Probabilistic Framework for Lifelong Test-Time Adaptationcited by 44
  34. 2021Neural Mean Discrepancy for Efficient Out-of-Distribution Detectioncited by 42
  35. 2023Feature Alignment and Uniformity for Test Time Adaptationcited by 42
  36. 2021PRIME: A few primitives can boost robustness to common corruptionscited by 40
  37. 2025A Survey of Test-Time Compute: From Intuitive Inference to Deliberate Reasoningcited by 40
  38. 2022ActMAD: Activation Matching to Align Distributions for Test-Time-Trainingcited by 38

50 most cited of 427 All 427 (JSON)