Masamsa/3d-print-anomaly-patchcore

3D Print Failure Detector — PatchCore

Anomaly detection model for detecting 3D print failures (spaghetti, detachment, etc.). Trained only on normal prints using PatchCore with resnet18 backbone.

Metrics

{
  "image_AUROC": 0.8632,
  "image_F1Score": 0.9248
}

Threshold

Auto-computed threshold: 17.3327
Frames with anomaly score > threshold are classified as failure.

Dataset

• Train: 1491 normal frames from YouTube printer timelapse videos
• Test: 373 normal + 850 failure frames

Inference

ONNX model: weights/onnx/model.onnx
Threshold config: weights/onnx/config.json