You need document layout analysis on pages that are scanned, photographed, faxed, microfilmed or otherwise damaged, and you are shipping a closed-source product.
Typical cases:
- Digitisation and archives. Microfilm, bound books, historical records.
- Legal e-discovery. Scanned exhibits, photocopies, faxes.
- Claims processing. Photographs of forms taken on a phone.
- Records management. Replacing a per-page cloud OCR bill with a client-side model.
AGPL-3.0 requires you to publish your source if you run a modified version as a network service. Most commercial products cannot accept that. The commercial licence removes it.
Against cloud OCR APIs. This runs on the client. No page ever leaves the device, which matters for legal, medical and financial documents, and there is no per-page cost.
Against larger layout models. 8.8 MB loads in a browser. A 300 MB transformer does not.
Against YOLOv8n-DocLayNet. Better on damaged pages by a factor of 2.3, detects form fields, and reports page condition. Worse on clean rendered pages, which is stated plainly rather than hidden.
No comparable detector reports these. Eight scalars per page: skew, blur, sensor noise, bleed-through, perspective warp, physical tearing, handwriting presence, and whether the page is a genuine digital document or a scan carrying an invisible OCR layer.
That last signal has an audit use. A scanned page with a synthetic text layer looks born-digital to most pipelines, and this model flags it.
Evaluate under AGPL first. The model, weights and examples are public, and no licence is needed to try them.
When you are ready to ship:
- contact@ginexys.com for pricing, evaluation and tier selection
- legal@ginexys.com for licence terms, redistribution and compliance
Ginexys, a d/b/a of Canworks, LLC. https://ginexys.com