Detect Image Anomaly (PatchCore)

@amazon1 VersionDetect

PatchCore anomaly detector: takes an Image and emits {score: Double, heatmap: Image} — an image-level anomaly score plus a per-pixel heatmap Image. Sharper localisation than PaDiM at higher cost from coreset memory-bank matching over backbone features.

How it fits

Typical backends

Batch review of captured parts: each stored image is scored and its heatmap streamed to a browser for an inspector to triage offline.

Inline reject decision: a video feed is scored, an expression stage applies the pass/fail threshold, and a filter forwards only the failing parts as JSON.

Caveats

Limitationmodel_path and device are fixed at startup; swapping the checkpoint or backend requires a redeploy.
FallbackRequesting a GPU when none is available falls back to CPU at startup, and an empty input Image short-circuits to a zero score and a 1x1 black heatmap without running the network.
AccuracyThe heatmap is min-max normalised per frame, so absolute heatmap intensities are not comparable across frames; compare the scalar score instead.
CostInference cost scales with the size of the coreset memory bank, so a larger bank slows per-frame matching proportionally.
Hard constraintThe checkpoint misbehaves on any product class it was not fitted on; there is no automatic out-of-distribution rejection.

Versions

145909b1latestdefault
Automated release
6/29/2026