paddleocr

PaddleOCR is a comprehensive OCR and document AI toolkit developed by PaddlePaddle. It converts PDF documents and images into structured, LLM-ready data in JSON or Markdown format. Key components: - PP-OCRv6: A multilingual text recognition model supporting 50+ languages in a single unified model, including Chinese, English, Japanese, and 46 Latin-script languages. It also covers Cyrillic, Arabic, Devanagari, and other scripts across 100+ languages total. Available in three size tiers (tiny 1.5M, small 7.7M, medium 34.5M parameters) for edge, mobile, and server deployment. - PaddleOCR-VL: A vision-language model (0.9B parameters) for document parsing. It achieves strong results on benchmarks like OmniDocBench and handles text, formulas, tables, charts, ancient documents, seals, and rare characters. Outputs structured Markdown and JSON with coordinate information. - PP-StructureV3: A structure-aware document conversion pipeline that provides fine-grained coordinate information for table cells, text regions, and other layout elements. - HPD-Parsing: A lightweight vision-language model for high-throughput document parsing with hierarchical parallel decoding. The toolkit supports multiple hardware backends including NVIDIA GPU, Intel CPU, Kunlunxin XPU, and various AI accelerators. It can be deployed via ONNX, OpenVINO, TensorRT, or served as an HTTP API. PaddleOCR is integrated with AI application platforms such as Dify, RAGFlow, Pathway, and Cherry Studio. An official browser inference SDK, PaddleOCR.js, allows running PP-OCR models directly in the browser. The project also supports converting Word, Excel, and PowerPoint documents to Markdown, and exporting parsed results to DOCX format.