NaviDC-OCR — document parsing, digital and camera-captured

A 1.2B document-parsing VLM that reads layout, text, tables, formulas and code off flat scans and photographed / crumpled pages, and returns Markdown.

model · paper · code

Document page

Parsing mode

Full page, boxes — digital pages & flat scans Full page, multi-point — photos, curved or crumpled pages Single region — the image is one table / formula / …

Examples from the NaviDC-OCR model card

Document page	Parsing mode	Region type