Defect Detection
Train AI to detect surface defects — scratches, dents, cracks, and anomalies — in your own manufacturing parts.
Surface defects are rare, product-specific, and hard to describe with a generic model. Off-the-shelf detectors miss the defects that matter to your line, and sending proprietary part imagery to a vendor cloud is often a non-starter.
Stamped or machined metal panels, castings, welds, PCBs, and coated surfaces — labeled with classes such as scratch, dent, crack, porosity, or contamination.
- 1Upload your own inspection images in the browser — no cloud upload is required to annotate.
- 2Mark defects with boxes or polygons, accelerated by on-device AI-assisted segmentation (Smart Select).
- 3Define your own defect classes and manage negative (defect-free) samples for balanced training.
- 4Export a validated YOLO or COCO dataset, then train on a GPU worker your organization controls.
What you get
A defect detector trained on your parts and built by your inspectors — not only ML engineers — with proprietary imagery kept inside your boundary.
Current readiness
Annotation, AI-assisted labeling, and validated export are live today. Model training runs on a GPU worker you control (private-worker architecture); hosted training is on the roadmap. Sector accuracy is validated per pilot.
What Anastig Studio brings to defect detection.
Statuses are live from our capability catalog — anything marked “Planned” is on the roadmap, not shipped.
Bounding Box
Available nowDraw and edit rectangular detection boxes in original image coordinates.
Polygon Annotation
Available nowOutline objects with precise polygon shapes for instance segmentation.
YOLO Detection Export
Available nowNormalized YOLO .txt labels + dataset.yaml (YOLO format).
Start in the product
Scope it with us
Tell us about your defect detection task — images, throughput, and where the model needs to run — and we'll come back with an honest read on what works today and what needs a pilot.
Contact us