Code and Supporting Materials for "Homologous Peer Matching for Training-Free Industrial Anomaly Detection: Benefits and Limits of Structured Self-Reference"
Description
This repository contains a minimal reference implementation and supporting materials for the study “Homologous Peer Matching for Training-Free Industrial Anomaly Detection: Benefits and Limits of Structured Self-Reference.” The released materials include implementations of unrestricted semantic matching (U), homologous exact matching (Exact), homologous multi-peer consensus (HMS), instance-level anomaly-score aggregation, and basic evaluation procedures. Only methods and parameters described in the associated manuscript are included. Raw industrial solder-ball images, manual anomaly annotations, instance-level industrial records, and other potentially confidential industrial information are not included. The VisA dataset used for external evaluation is publicly available from its original source and is not redistributed in this repository. The released package is intended to support inspection and reuse of the methodology rather than to provide complete reproduction of the private industrial dataset experiments.
Files
Steps to reproduce
Install the required Python packages listed in requirements.txt. Run demo.py to execute the synthetic demonstration. The core implementations of U, Exact, and HMS are provided in the src directory. Private industrial images and manual annotations are not included in this release.
Institutions
- North University of ChinaShanxi, Taiyuan