mlsanityData to 7 Oct 2026

Data Invariants to Understand Unsupervised Out-of-Distribution Detection

arXiv 2111.13362 ↗First posted 26 Nov 2021cs.CVcited by 5

Not on the map. Why?

Abstract (arXiv)

Unsupervised out-of-distribution (U-OOD) detection has recently attracted much attention due its importance in mission-critical systems and broader applicability over its supervised counterpart. Despite this increase in attention, U-OOD methods suffer from important shortcomings. By performing a large-scale evaluation on different benchmarks and image modalities, we show in this work that most popular state-of-the-art methods are unable to consistently outperform a simple anomaly detector based on pre-trained features and the Mahalanobis distance (MahaAD). A key reason for the inconsistencies of these methods is the lack of a formal description of U-OOD. Motivated by a simple thought experiment, we propose a characterization of U-OOD based on the invariants of the training dataset. We show how this characterization is unknowingly embodied in the top-scoring MahaAD method, thereby explaining its quality. Furthermore, our approach can be used to interpret predictions of U-OOD detectors and provides insights into good practices for evaluating future U-OOD methods.

Key facts. Cited by 5 papers in the map (data through 7 Oct 2026). Builds on 64 of its 101 references that are in the map. First posted on arXiv 26 Nov 2021.

Builds on

64 of 101 references.

References · 64

  1. 2016A Baseline for Detecting Misclassified and Out-of-Distribution Examples in Neural Networkscited by 2,362
  2. 2016Concrete Problems in AI Safetycited by 1,582
  3. 2018Glow: Generative Flow with Invertible 1x1 Convolutionscited by 1,539
  4. 2018A Simple Unified Framework for Detecting Out-of-Distribution Samples and Adversarial Attackscited by 1,472
  5. 2018Moment Matching for Multi-Source Domain Adaptationcited by 1,327
  6. 2017Enhancing The Reliability of Out-of-distribution Image Detection in Neural Networkscited by 1,325
  7. 2018Learning deep representations by mutual information estimation and maximizationcited by 1,265
  8. 2021Generalized Out-of-Distribution Detection: A Surveycited by 796
  9. 2020PaDiM: a Patch Distribution Modeling Framework for Anomaly Detection and Localizationcited by 591
  10. 2017Unsupervised Anomaly Detection with Generative Adversarial Networks to Guide Marker Discoverycited by 566
  11. 2019Memorizing Normality to Detect Anomaly: Memory-augmented Deep Autoencoder for Unsupervised Anomaly Detectioncited by 560
  12. 2021CutPaste: Self-Supervised Learning for Anomaly Detection and Localizationcited by 529
38 more
  1. 2019Using Self-Supervised Learning Can Improve Model Robustness and Uncertaintycited by 511
  2. 2019Likelihood Ratios for Out-of-Distribution Detectioncited by 451
  3. 2018Do Deep Generative Models Know What They Don't Know?cited by 443
  4. 2020CSI: Novelty Detection via Contrastive Learning on Distributionally Shifted Instancescited by 441
  5. 2020A Unifying Review of Deep and Shallow Anomaly Detectioncited by 423
  6. 2017PixelCNN++: Improving the PixelCNN with Discretized Logistic Mixture Likelihood and Other Modificationscited by 415
  7. 2018GANomaly: Semi-Supervised Anomaly Detection via Adversarial Trainingcited by 409
  8. 2020Generalized ODIN: Detecting Out-of-distribution Image without Learning from Out-of-distribution Datacited by 408
  9. 2019Uninformed Students: Student-Teacher Anomaly Detection with Discriminative Latent Embeddingscited by 386
  10. 2019Deep Learning for Anomaly Detection: A Surveycited by 357
  11. 2021CFLOW-AD: Real-Time Unsupervised Anomaly Detection with Localization via Conditional Normalizing Flowscited by 320
  12. 2020Sub-Image Anomaly Detection with Deep Pyramid Correspondencescited by 297
  13. 2020Multiresolution Knowledge Distillation for Anomaly Detectioncited by 285
  14. 2018Deep Anomaly Detection Using Geometric Transformationscited by 269
  15. 2021SSD: A Unified Framework for Self-Supervised Outlier Detectioncited by 264
  16. 2018Improving Unsupervised Defect Segmentation by Applying Structural Similarity to Autoencoderscited by 263
  17. 2020Classification-Based Anomaly Detection for General Datacited by 210
  18. 2019OCGAN: One-class Novelty Detection Using GANs with Constrained Latent Representationscited by 199
  19. 2019Input complexity and out-of-distribution detection with likelihood-based generative modelscited by 186
  20. 2020PANDA: Adapting Pretrained Features for Anomaly Detection and Segmentationcited by 175
  21. 2018Latent Space Autoregression for Novelty Detectioncited by 149
  22. 2020Likelihood Regret: An Out-of-Distribution Detection Score For Variational Auto-encodercited by 145
  23. 2020Learning and Evaluating Representations for Deep One-class Classificationcited by 133
  24. 2020DROCC: Deep Robust One-Class Classificationcited by 121
  25. 2020Modeling the Distribution of Normal Data in Pre-Trained Deep Features for Anomaly Detectioncited by 115
  26. 2016Deep Structured Energy Based Models for Anomaly Detectioncited by 113
  27. 2020Deep Nearest Neighbor Anomaly Detectioncited by 103
  28. 2020VAEBM: A Symbiosis between Variational Autoencoders and Energy-based Modelscited by 103
  29. 2021Mean-Shifted Contrastive Loss for Anomaly Detectioncited by 87
  30. 2021Divide-and-Assemble: Learning Block-wise Memory for Unsupervised Anomaly Detectioncited by 83
  31. 2018Anomaly Detection using One-Class Neural Networkscited by 82
  32. 2020Density of States Estimation for Out-of-Distribution Detectioncited by 74
  33. 2021Revisiting Mahalanobis Distance for Transformer-Based Out-of-Domain Detectioncited by 67
  34. 2020Understanding Anomaly Detection with Deep Invertible Networks through Hierarchies of Distributions and Featurescited by 64
  35. 2021Hierarchical VAEs Know What They Don't Knowcited by 58
  36. 2019Detecting semantic anomaliescited by 50
  37. 2020Iterative energy-based projection on a normal data manifold for anomaly localizationcited by 47
  38. 2020Improved anomaly detection by training an autoencoder with skip connections on images corrupted with Stain-shaped noisecited by 39

50 most cited of 64 All 64 (JSON)

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2023: 2; 2024: 2; 2025: 0; 2026: 1.

5 papers.

  1. 2023No Free Lunch: The Hazards of Over-Expressive Representations in Anomaly Detectioncited by 7
  2. 2023Unsupervised out-of-distribution detection for safer robotically guided retinal microsurgerycited by 1
  3. 2024Iterative Deployment Exposure for Unsupervised Out-of-Distribution Detectioncited by 0
  4. 2024Learning Non-Linear Invariants for Unsupervised Out-of-Distribution Detectioncited by 0
  5. 2026Modality-Aware Out-of-Distribution Detection for Multi-Modal Action Recognitioncited by 0