PatchCore Anomaly Detection

A from-scratch PatchCore implementation for unsupervised industrial anomaly detection, evaluated at both image and pixel level.

Explore the implementation and benchmark details on GitHub.

Problem

Industrial inspection must identify defects even when labeled examples of every defect type are unavailable. Unsupervised anomaly detection addresses this by learning from normal examples.

Approach

The project implements PatchCore from scratch and evaluates both whether an image is anomalous and where anomalies occur within it. This connects image-level inspection with pixel-level defect localization.

Engineering Focus

  • Unsupervised industrial anomaly detection
  • From-scratch algorithm implementation
  • Image-level anomaly detection
  • Pixel-level anomaly localization
  • Benchmark evaluation

Related Knowledge

See the PatchCore knowledge page for the algorithm's concepts and the MVTec AD dataset page for industrial inspection data.