Universal Transformation of One-Class Classifiers for Unsupervised Anomaly Detection

2026-02-13 · Source: Computer Vision and Pattern Recognition · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Computer Vision · Depth: Expert, quick

Summary

A new dataset folding method transforms existing one-class classifier-based anomaly detectors into fully unsupervised methods for image and video anomaly detection. This approach operates under the weak assumptions that anomalies are uncommon and heterogeneous within the training dataset, allowing multiple independently trained one-class classifiers to filter the training data. The transformation requires no modifications to the underlying anomaly detector, only algorithmic selection of data subsets for training. This method creates the first unsupervised logical anomaly detectors and achieves state-of-the-art performance on the MVTec AD, ViSA, and MVTec Loco AD datasets, effectively linking improvements in one-class classifiers to the unsupervised domain.

Key takeaway

For Computer Vision Engineers developing anomaly detection systems, this method offers a path to convert existing one-class classifiers into unsupervised detectors. You can achieve state-of-the-art performance on datasets like MVTec AD and ViSA without re-architecting your core models, simply by implementing the proposed data subset selection and filtering strategy. This allows you to leverage advancements in one-class classification directly for unsupervised tasks.

Key insights

A dataset folding method converts one-class classifiers into unsupervised anomaly detectors without modifying their core.

Principles

• Anomalies are uncommon in training data.
• Anomalies are generally heterogeneous.

Method

Multiple independently trained one-class classifiers filter training data subsets, enabling unsupervised anomaly detection without altering the base detector's architecture.

In practice

• Apply to existing one-class image detectors.
• Extend to video anomaly detection tasks.

Topics

• Unsupervised Anomaly Detection
• One-Class Classification
• Dataset Folding Method
• Computer Vision
• MVTec AD

Best for: Computer Vision Engineer, Research Scientist, AI Researcher, AI Scientist, Machine Learning Engineer

Related on AIssential

• A Comprehensive, Practical Guide to Unsupervised Anomaly Detection — Unsupervised anomaly detection models normal behavior to identify rare, unknown deviations.
• Decomposing one-class support vector machine into an ensemble of one-data support vector machines — Decomposing OCSVM into an ensemble of one-data models significantly accelerates processing for large datasets.
• Rethinking Structural Anomaly Detection: From Decision Boundaries to Projection Operators — Structural anomaly detection improves by projecting data onto a normal manifold, identifying anomalies via projection residuals.
• Encoding Categorical Data for Outlier Detection — Effectively encoding categorical data into numeric formats is crucial for most unsupervised outlier detection algorithms.
• MambaADv2: Evolving Duality-enhanced State Space Model for Unsupervised Anomaly Detection — MambaADv2 employs duality-enhanced state space models for efficient, multi-class unsupervised anomaly detection, addressing CNN and Transformer limitations.
• Conformal Anomaly Detection in Python: Moving Beyond Heuristic Thresholds with 'nonconform' — Conformal anomaly detection converts raw anomaly scores into statistically calibrated p-values, eliminating heuristic thresholding.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computer Vision and Pattern Recognition.