Pre-process
Resize, bilateral-filter denoise, grayscale conversion.
Live · Manufacturing
Visual defect inspection — running live on this server.
Pick one of the three synthetic samples below, or drag-drop your own image. The pipeline runs in ~25 ms on this CPU and highlights pits, scratches, and contamination with bounding boxes and confidence scores.
1 · Pick a sample
Loading samples…
2 · Or upload your own
Drag & drop an image, or click below. JPEG / PNG / WebP, max 5 MB.
No data is stored. Rate limit: 30 inspections per minute per IP.
Inspection result
Pick a sample or upload an image
Results appear here as a side-by-side annotated image with the defect list and timing.
How it works
No proprietary models or licensed datasets. Everything below is open source and CPU-only.
Estimate background
Median blur with a large kernel captures the dominant surface.
Residual
Absolute difference between original and background highlights local anomalies.
Threshold
Otsu (or fallback statistical cut) binarises the residual into candidate anomaly regions.
Cleanup
Morphological open/close removes single-pixel noise and merges nearby fragments.
Classify
Connected components are scored by aspect ratio and extent into scratch / pit / contaminant / anomaly.
What this demo proves
- · Real ML pipeline on real images, in your browser, in ~25 ms
- · No GPU required — runs on a 2-core CPU
- · Fully open-source (OpenCV, scikit-image, FastAPI)
- · Classification is explainable (rule-based on geometric features)
- · Plug a deep-learning model in the same place when accuracy demands it
- · Same deployment pattern works for ten verticals
Want this on your shop floor, with your defects?
We build production visual-inspection systems with your data, your defect classes, and your tolerance for false positives — typically 6–8 weeks from kickoff to running on real product.
Ready to start
Turn one AI use case into measurable production value.
Book a 30-minute consultation. We will walk through the use case, sketch the value case, and tell you honestly whether we can help.