PaddleOCR

Overview

PaddleOCR is an OCR toolkit and document AI engine that converts PDFs and images into structured, LLM-ready data. It provides high-accuracy text recognition, document parsing, and multilingual support for building intelligent RAG and Agentic applications.

Key Features

• Intelligent document parsing converts complex PDFs and images into Markdown or JSON, with structure-aware conversion and fine-grained coordinates.
• PaddleOCR-VL and PP-StructureV3 support text, formula, table, chart, seal, and ancient document recognition.
• Universal text recognition supports 100+ languages, with PP-OCRv6 covering 50 languages in a single model without model switching.
• PP-OCRv6 improves detection by 4.6% and recognition by 5.1% over PP-OCRv5, with 5.2× CPU inference speedup end-to-end.
• The developer ecosystem integrates with Dify, RAGFlow, Pathway, and Cherry Studio and supports building LLM data pipelines.
• One-click deployment supports NVIDIA GPU, Intel CPU, Kunlunxin XPU, and diverse AI accelerators.

Use Cases

• Building RAG pipelines that need accurate parsing of PDFs, images, and complex documents for LLM ingestion.
• Converting scanned documents, tables, formulas, and charts into Markdown or JSON for structured data workflows.
• Multilingual document processing across 100+ languages, including Chinese, English, Japanese, Korean, Arabic, and other scripts.
• Embedding OCR into applications through Python, C++, C#, Java, or browser-based inference.

Getting Started

• Try the online Experience Center and APIs on the official website without setup.
• For local deployment, follow the PP-OCR, PaddleOCR-VL, or PP-StructureV3 documentation based on your needs.

Deployment & Requirements

• Python 3.8–3.12; OS support for Linux, Windows, and macOS; hardware support for CPU, GPU, XPU, and NPU.
• One-click deployment supports NVIDIA GPU, Intel CPU, Kunlunxin XPU, and diverse AI accelerators.
• High-performance inference supports CUDA 12 and can use Paddle Inference or ONNX Runtime backends.

Before You Adopt

• License: Apache-2.0. Review its terms before using, modifying, or distributing the project.
• Basic text recognition requires only minimal core dependencies, while document parsing and information extraction require optional dependencies.
• High-performance inference depends on CUDA 12 and can use Paddle Inference or ONNX Runtime backends.
• Full support is noted for PaddlePaddle framework versions 3.1.0 and 3.1.1.