PaddleOCR

Backed by Baidu's PaddlePaddle deep learning framework, PaddleOCR is an Apache 2.0 licensed toolkit for optical character recognition and document parsing that converts images and PDFs into structured JSON or Markdown. Its PP-OCR model series handles text detection and recognition across more than 80 languages, while PP-StructureV3 performs layout analysis, table recognition, and formula recognition with fine-grained coordinates, and PaddleOCR-VL adds a lightweight 0.9B parameter vision-language model for complex document understanding covering handwriting, seals, and charts. Ultra-light model variants target mobile and edge deployment. The toolkit is widely embedded in other software: projects such as RAGFlow, Dify, and Microsoft's OmniParser rely on it for text extraction, and the repository counts tens of thousands of GitHub stars with thousands of dependent projects. Typical users are developers building document pipelines and RAG systems that need accurate self-hosted OCR.