# [Anomaly Detection] PatchCore on Orange line 🍊 — retained record

> Retained reddit record from reddit:computervision.

Canonical: https://highsignal.app/data/records/1bdaa71baf796269

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:computervision)  published 2026-10-28 20:08 ingested 2026-10-29 02:37

[original source ↗](https://www.reddit.com/r/computervision/comments/1xhz195/anomaly_detection_patchcore_on_orange_line/)

## retained text



```
PatchCore running on orange line, trained on good oranges only 🍊

* RF-DETR (instance seg) detects every orange, ByteTrack tracks them
* Cropped orange imgs are sent to PatchCore, which compares each patch to a memory bank of normal patches → heatmap 
* Each orange is inspected multiple times while it rolls around on rolling conveyor, as anomaly could be present only on one side and not the other

This is super useful in cases where anomalous data is hard to come by, and anomaly detection training requires zero defect/anomalous labels. PatchCore runs at \~100ms on Jetson Orin AGX (at \~200 images, compute time scales with training data).
```



document fetched 2026-10-29 00:24 · canonical https://reddit.com/r/computervision/comments/1xhz195/anomaly_detection_patchcore_on_orange_line
