Universal Transformation of One-Class Classifiers for Unsupervised Anomaly Detection
Declan McIntosh, Alexandra Branzan Albu · 2026
Read Universal Transformation of One-Class Classifiers for Unsupervised Anomaly Detection by Declan McIntosh, Alexandra Branzan Albu (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-13091.txt.
Description
Anomaly detection in images and video is usually framed as one-class classification: train only on normal examples. That assumption leaves methods vulnerable to label noise in training data. Declan McIntosh and Alexandra Branzan Albu present Bootstrap Aggregation Anomaly Filtering, which transforms any one-class anomaly detector into a fully unsupervised method. The trick exploits two traits of anomalies: they are rare and heterogeneous.
Several separately trained instances of supervised one-class classifiers sift the training set for anomalies, using modified bootstrap aggregation. No change to the underlying detector is needed; only the data bags used for training change. The method achieves state-of-the-art fully unsupervised results on MVTec AD, ViSA, and MVTec LOCO AD, including unsupervised logical anomaly detectors for images, a first for the field. Improvements to one-class classifiers transfer directly to the unsupervised domain.
Catalog record
| CREATOR | Declan McIntosh, Alexandra Branzan Albu |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 45 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-13091