CLAD: A Contrastive Learning based Method for Multi-Class Anomaly Detection

Runtang Guo, Hong Li, Tongfei Chen, Yuguang Yang, Peng Zhou, Linlin Yang, Guodong Guo, Baochang Zhang

Keywords: Industrial anomaly detection, Multi-class anomaly detection, Contrastive Learning

TL;DR: A Contrastive Learning based Method for Multi-Class Anomaly Detection

Abstract:

Anomaly detection is crucial yet challenging in industrial production, especially in multi-class scenarios. Existing high-performance unsupervised methods often suffer from low efficiency and high model complexity. While lightweight discriminator-based detectors have been proposed, they are typically designed for single-class detection and exhibit significant performance degradation when extended to multi-class tasks. To address these limitations, we propose a novel Contrastive Learning-based multi-class Anomaly Detection (CLAD) method.
Our approach first encodes multi-class normal images to generate normal samples in the feature space, then synthesizes anomalous samples in this encoded space. We then employ an Adapter network to compress the samples and leverage contrastive learning to effectively cluster normal and anomalous samples across multiple classes. Finally, a discriminator network is used for anomaly classification and identification. By leveraging anomaly sample generation and a two-stage training process, our framework achieves state-of-the-art performance on the MVTec and VisA datasets under the discriminator-based paradigm. Our key contributions include a novel framework for multi-class anomaly detection, efficient sample generation techniques, and a comprehensive evaluation of model configurations.

Primary Area: applications to computer vision, audio, language, and other modalities

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 10576