Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Self-Supervised Anomaly Detection in Computer Vision and Beyond: A Survey and Outlook

Hadi Hojjati, Thi Kieu Khanh Ho, Narges Armanfard

TL;DR

The paper surveys self-supervised anomaly detection (SSL-AD), addressing the problem of identifying deviations from normal behavior when labeled anomalies are scarce. It classifies SSL-AD methods into self-predictive and contrastive approaches, discusses problem formulations (one-class vs multi-class), anomaly scoring, and performance benchmarks, and extends analysis to non-image domains (audio, time-series, graphs, tabular/text data). Key contributions include a taxonomy of proxy tasks (e.g., GOAD, GEOM, CutPaste, NSA), an overview of scoring strategies, and a comparative discussion of SSL-AD performance on standard datasets like CIFAR-10 and MVTecAD, along with future directions such as negative sampling, multi-modal fusion, and efficient learning. The findings underscore SSL as a leading paradigm for AD, capable of leveraging large amounts of unlabeled data to learn robust normal representations and enabling scalable, cross-domain anomaly detection with practical impact across diverse sectors.

Abstract

Anomaly detection (AD) plays a crucial role in various domains, including cybersecurity, finance, and healthcare, by identifying patterns or events that deviate from normal behaviour. In recent years, significant progress has been made in this field due to the remarkable growth of deep learning models. Notably, the advent of self-supervised learning has sparked the development of novel AD algorithms that outperform the existing state-of-the-art approaches by a considerable margin. This paper aims to provide a comprehensive review of the current methodologies in self-supervised anomaly detection. We present technical details of the standard methods and discuss their strengths and drawbacks. We also compare the performance of these models against each other and other state-of-the-art anomaly detection models. Finally, the paper concludes with a discussion of future directions for self-supervised anomaly detection, including the development of more effective and efficient algorithms and the integration of these techniques with other related fields, such as multi-modal learning.

Self-Supervised Anomaly Detection in Computer Vision and Beyond: A Survey and Outlook

TL;DR

The paper surveys self-supervised anomaly detection (SSL-AD), addressing the problem of identifying deviations from normal behavior when labeled anomalies are scarce. It classifies SSL-AD methods into self-predictive and contrastive approaches, discusses problem formulations (one-class vs multi-class), anomaly scoring, and performance benchmarks, and extends analysis to non-image domains (audio, time-series, graphs, tabular/text data). Key contributions include a taxonomy of proxy tasks (e.g., GOAD, GEOM, CutPaste, NSA), an overview of scoring strategies, and a comparative discussion of SSL-AD performance on standard datasets like CIFAR-10 and MVTecAD, along with future directions such as negative sampling, multi-modal fusion, and efficient learning. The findings underscore SSL as a leading paradigm for AD, capable of leveraging large amounts of unlabeled data to learn robust normal representations and enabling scalable, cross-domain anomaly detection with practical impact across diverse sectors.

Abstract

Anomaly detection (AD) plays a crucial role in various domains, including cybersecurity, finance, and healthcare, by identifying patterns or events that deviate from normal behaviour. In recent years, significant progress has been made in this field due to the remarkable growth of deep learning models. Notably, the advent of self-supervised learning has sparked the development of novel AD algorithms that outperform the existing state-of-the-art approaches by a considerable margin. This paper aims to provide a comprehensive review of the current methodologies in self-supervised anomaly detection. We present technical details of the standard methods and discuss their strengths and drawbacks. We also compare the performance of these models against each other and other state-of-the-art anomaly detection models. Finally, the paper concludes with a discussion of future directions for self-supervised anomaly detection, including the development of more effective and efficient algorithms and the integration of these techniques with other related fields, such as multi-modal learning.
Paper Structure (26 sections, 21 equations, 4 figures, 5 tables)

This paper contains 26 sections, 21 equations, 4 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Anomaly Detection: Terminology and Common Practices
  4. Anomaly, Outlier, Novelty, Out-of-Distribution Detection
  5. Types of Anomalies
  6. Availability of Data Labels
  7. Self-supervised Learning for Anomaly Detection
  8. Problem Formulation
  9. Algorithms
  10. Anomaly Scoring
  11. Performance Evaluation
  12. Self-Predictive Methods in Anomaly Detection
  13. Contrastive Methods
  14. Self-Supervised Anomaly Detection Beyond Images
  15. Audio Anomaly Detection
  16. ...and 11 more sections

Figures (4)

  • Figure 1: Normal samples are shown in green, anomalies in red, outliers in blue and novelties in purple. The dataset of animals is denoted by a light-blue dashed box while a dashed dark-red box shows other out-of-distribution datasets.
  • Figure 2: Several examples of pseudo-label generation processes that are associated with two main categories of SSL-AD. $x$ is the pseudo-labeled input and $f_{\theta}$ is the feature extractor.
  • Figure 3: Performance of anomaly detection algorithms on the MVTecAD dataset. Each group of algorithm is denoted by a different colour.
  • Figure 4: Timeline of Self-Supervised Anomaly Detection Papers. Papers concerning the algorithms are distinguished from the application papers. The category of each algorithm is denoted by a distinctive color.