Nanonets OCR
lastmod 2025-06-18created 2025-06-181 edits
An OCR model that converts documents into structured Markdown
About
- A high-performance image-to-Markdown OCR model that goes beyond simple character recognition to convert entire documents into Markdown structure
- Converts math equations to LaTeX, adds automatic descriptions to images, and outputs tables as HTML/Markdown tables, producing output optimized for LLM use
- Excels at handling individual document components, e.g. recognizing signatures, watermarks, and checkboxes and converting them to <signature>, <watermark>, and ☐/☑
- Easy to use via Hugging Face Transformers or a vLLM server, and also available as a web app via the docext library
- Very high accuracy and level of structure across diverse document types and complex layouts, making it very useful for contracts, forms, reports, and more