[Anomaly Detection] PatchCore on Orange line 🍊

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, comput