[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"similar-PaddlePaddle--PaddleOCR":3,"tool-PaddlePaddle--PaddleOCR":64},[4,17,27,35,43,56],{"id":5,"name":6,"github_repo":7,"description_zh":8,"stars":9,"difficulty_score":10,"last_commit_at":11,"category_tags":12,"status":16},3808,"stable-diffusion-webui","AUTOMATIC1111\u002Fstable-diffusion-webui","stable-diffusion-webui 是一个基于 Gradio 构建的网页版操作界面，旨在让用户能够轻松地在本地运行和使用强大的 Stable Diffusion 图像生成模型。它解决了原始模型依赖命令行、操作门槛高且功能分散的痛点，将复杂的 AI 绘图流程整合进一个直观易用的图形化平台。\n\n无论是希望快速上手的普通创作者、需要精细控制画面细节的设计师，还是想要深入探索模型潜力的开发者与研究人员，都能从中获益。其核心亮点在于极高的功能丰富度：不仅支持文生图、图生图、局部重绘（Inpainting）和外绘（Outpainting）等基础模式，还独创了注意力机制调整、提示词矩阵、负向提示词以及“高清修复”等高级功能。此外，它内置了 GFPGAN 和 CodeFormer 等人脸修复工具，支持多种神经网络放大算法，并允许用户通过插件系统无限扩展能力。即使是显存有限的设备，stable-diffusion-webui 也提供了相应的优化选项，让高质量的 AI 艺术创作变得触手可及。",162132,3,"2026-04-05T11:01:52",[13,14,15],"开发框架","图像","Agent","ready",{"id":18,"name":19,"github_repo":20,"description_zh":21,"stars":22,"difficulty_score":23,"last_commit_at":24,"category_tags":25,"status":16},1381,"everything-claude-code","affaan-m\u002Feverything-claude-code","everything-claude-code 是一套专为 AI 编程助手（如 Claude Code、Codex、Cursor 等）打造的高性能优化系统。它不仅仅是一组配置文件，而是一个经过长期实战打磨的完整框架，旨在解决 AI 代理在实际开发中面临的效率低下、记忆丢失、安全隐患及缺乏持续学习能力等核心痛点。\n\n通过引入技能模块化、直觉增强、记忆持久化机制以及内置的安全扫描功能，everything-claude-code 能显著提升 AI 在复杂任务中的表现，帮助开发者构建更稳定、更智能的生产级 AI 代理。其独特的“研究优先”开发理念和针对 Token 消耗的优化策略，使得模型响应更快、成本更低，同时有效防御潜在的攻击向量。\n\n这套工具特别适合软件开发者、AI 研究人员以及希望深度定制 AI 工作流的技术团队使用。无论您是在构建大型代码库，还是需要 AI 协助进行安全审计与自动化测试，everything-claude-code 都能提供强大的底层支持。作为一个曾荣获 Anthropic 黑客大奖的开源项目，它融合了多语言支持与丰富的实战钩子（hooks），让 AI 真正成长为懂上",138956,2,"2026-04-05T11:33:21",[13,15,26],"语言模型",{"id":28,"name":29,"github_repo":30,"description_zh":31,"stars":32,"difficulty_score":23,"last_commit_at":33,"category_tags":34,"status":16},2271,"ComfyUI","Comfy-Org\u002FComfyUI","ComfyUI 是一款功能强大且高度模块化的视觉 AI 引擎，专为设计和执行复杂的 Stable Diffusion 图像生成流程而打造。它摒弃了传统的代码编写模式，采用直观的节点式流程图界面，让用户通过连接不同的功能模块即可构建个性化的生成管线。\n\n这一设计巧妙解决了高级 AI 绘图工作流配置复杂、灵活性不足的痛点。用户无需具备编程背景，也能自由组合模型、调整参数并实时预览效果，轻松实现从基础文生图到多步骤高清修复等各类复杂任务。ComfyUI 拥有极佳的兼容性，不仅支持 Windows、macOS 和 Linux 全平台，还广泛适配 NVIDIA、AMD、Intel 及苹果 Silicon 等多种硬件架构，并率先支持 SDXL、Flux、SD3 等前沿模型。\n\n无论是希望深入探索算法潜力的研究人员和开发者，还是追求极致创作自由度的设计师与资深 AI 绘画爱好者，ComfyUI 都能提供强大的支持。其独特的模块化架构允许社区不断扩展新功能，使其成为当前最灵活、生态最丰富的开源扩散模型工具之一，帮助用户将创意高效转化为现实。",107662,"2026-04-03T11:11:01",[13,14,15],{"id":36,"name":37,"github_repo":38,"description_zh":39,"stars":40,"difficulty_score":23,"last_commit_at":41,"category_tags":42,"status":16},3704,"NextChat","ChatGPTNextWeb\u002FNextChat","NextChat 是一款轻量且极速的 AI 助手，旨在为用户提供流畅、跨平台的大模型交互体验。它完美解决了用户在多设备间切换时难以保持对话连续性，以及面对众多 AI 模型不知如何统一管理的痛点。无论是日常办公、学习辅助还是创意激发，NextChat 都能让用户随时随地通过网页、iOS、Android、Windows、MacOS 或 Linux 端无缝接入智能服务。\n\n这款工具非常适合普通用户、学生、职场人士以及需要私有化部署的企业团队使用。对于开发者而言，它也提供了便捷的自托管方案，支持一键部署到 Vercel 或 Zeabur 等平台。\n\nNextChat 的核心亮点在于其广泛的模型兼容性，原生支持 Claude、DeepSeek、GPT-4 及 Gemini Pro 等主流大模型，让用户在一个界面即可自由切换不同 AI 能力。此外，它还率先支持 MCP（Model Context Protocol）协议，增强了上下文处理能力。针对企业用户，NextChat 提供专业版解决方案，具备品牌定制、细粒度权限控制、内部知识库整合及安全审计等功能，满足公司对数据隐私和个性化管理的高标准要求。",87618,"2026-04-05T07:20:52",[13,26],{"id":44,"name":45,"github_repo":46,"description_zh":47,"stars":48,"difficulty_score":23,"last_commit_at":49,"category_tags":50,"status":16},2268,"ML-For-Beginners","microsoft\u002FML-For-Beginners","ML-For-Beginners 是由微软推出的一套系统化机器学习入门课程，旨在帮助零基础用户轻松掌握经典机器学习知识。这套课程将学习路径规划为 12 周，包含 26 节精炼课程和 52 道配套测验，内容涵盖从基础概念到实际应用的完整流程，有效解决了初学者面对庞大知识体系时无从下手、缺乏结构化指导的痛点。\n\n无论是希望转型的开发者、需要补充算法背景的研究人员，还是对人工智能充满好奇的普通爱好者，都能从中受益。课程不仅提供了清晰的理论讲解，还强调动手实践，让用户在循序渐进中建立扎实的技能基础。其独特的亮点在于强大的多语言支持，通过自动化机制提供了包括简体中文在内的 50 多种语言版本，极大地降低了全球不同背景用户的学习门槛。此外，项目采用开源协作模式，社区活跃且内容持续更新，确保学习者能获取前沿且准确的技术资讯。如果你正寻找一条清晰、友好且专业的机器学习入门之路，ML-For-Beginners 将是理想的起点。",84991,"2026-04-05T10:45:23",[14,51,52,53,15,54,26,13,55],"数据工具","视频","插件","其他","音频",{"id":57,"name":58,"github_repo":59,"description_zh":60,"stars":61,"difficulty_score":10,"last_commit_at":62,"category_tags":63,"status":16},3128,"ragflow","infiniflow\u002Fragflow","RAGFlow 是一款领先的开源检索增强生成（RAG）引擎，旨在为大语言模型构建更精准、可靠的上下文层。它巧妙地将前沿的 RAG 技术与智能体（Agent）能力相结合，不仅支持从各类文档中高效提取知识，还能让模型基于这些知识进行逻辑推理和任务执行。\n\n在大模型应用中，幻觉问题和知识滞后是常见痛点。RAGFlow 通过深度解析复杂文档结构（如表格、图表及混合排版），显著提升了信息检索的准确度，从而有效减少模型“胡编乱造”的现象，确保回答既有据可依又具备时效性。其内置的智能体机制更进一步，使系统不仅能回答问题，还能自主规划步骤解决复杂问题。\n\n这款工具特别适合开发者、企业技术团队以及 AI 研究人员使用。无论是希望快速搭建私有知识库问答系统，还是致力于探索大模型在垂直领域落地的创新者，都能从中受益。RAGFlow 提供了可视化的工作流编排界面和灵活的 API 接口，既降低了非算法背景用户的上手门槛，也满足了专业开发者对系统深度定制的需求。作为基于 Apache 2.0 协议开源的项目，它正成为连接通用大模型与行业专有知识之间的重要桥梁。",77062,"2026-04-04T04:44:48",[15,14,13,26,54],{"id":65,"github_repo":66,"name":67,"description_en":68,"description_zh":69,"ai_summary_zh":69,"readme_en":70,"readme_zh":71,"quickstart_zh":72,"use_case_zh":73,"hero_image_url":74,"owner_login":75,"owner_name":75,"owner_avatar_url":76,"owner_bio":77,"owner_company":78,"owner_location":78,"owner_email":78,"owner_twitter":78,"owner_website":79,"owner_url":80,"languages":81,"stars":120,"forks":121,"last_commit_at":122,"license":123,"difficulty_score":10,"env_os":124,"env_gpu":125,"env_ram":126,"env_deps":127,"category_tags":132,"github_topics":133,"view_count":147,"oss_zip_url":78,"oss_zip_packed_at":78,"status":16,"created_at":148,"updated_at":149,"faqs":150,"releases":176},519,"PaddlePaddle\u002FPaddleOCR","PaddleOCR","Turn any PDF or image document into structured data for your AI. A powerful, lightweight OCR toolkit that bridges the gap between images\u002FPDFs and LLMs. Supports 100+ languages.","PaddleOCR 是一款基于百度飞桨框架开发的高性能开源光学字符识别工具包。它的核心能力是将图片、PDF 等文档中的文字提取出来，转换成计算机可读取的结构化数据，让机器真正“看懂”图文内容。\n\n面对海量纸质或电子文档，PaddleOCR 解决了人工录入效率低、数字化成本高的问题。尤其在人工智能领域，它扮演着连接图像与大型语言模型（LLM）的桥梁角色，能将视觉信息直接转化为文本输入，助力智能问答、文档分析等应用场景落地。\n\nPaddleOCR 适合开发者、算法研究人员以及有文档自动化需求的普通用户。其技术优势十分明显：不仅支持全球 100 多种语言的识别，还能在 Windows、Linux、macOS 等多个系统上运行，并灵活适配 CPU、GPU、NPU 等各类硬件。作为一个轻量级且社区活跃的开源项目，PaddleOCR 既能满足快速集成的需求，也能支撑前沿的视觉语言研究，是处理文字识别任务的理想选择。","\u003Cdiv align=\"center\">\n  \u003Cp>\n      \u003Cimg width=\"100%\" src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FPaddlePaddle_PaddleOCR_readme_f1dc4ea647a1.png\" alt=\"PaddleOCR Banner\">\n  \u003C\u002Fp>\n\nEnglish | [简体中文](.\u002Freadme\u002FREADME_cn.md) | [繁體中文](.\u002Freadme\u002FREADME_tcn.md) | [日本語](.\u002Freadme\u002FREADME_ja.md) | [한국어](.\u002Freadme\u002FREADME_ko.md) | [Français](.\u002Freadme\u002FREADME_fr.md) | [Русский](.\u002Freadme\u002FREADME_ru.md) | [Español](.\u002Freadme\u002FREADME_es.md) | [العربية](.\u002Freadme\u002FREADME_ar.md)\n\n\u003C!-- icon -->\n[![stars](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002FPaddlePaddle\u002FPaddleOCR?color=ccf)](https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR)\n[![forks](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fforks\u002FPaddlePaddle\u002FPaddleOCR.svg)](https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR)\n[![arXiv](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FPaddleOCR_3.0-Technical%20Report-b31b1b.svg?logo=arXiv)](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2507.05595)\n[![arXiv](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FPaddleOCR--VL-Technical%20Report-b31b1b.svg?logo=arXiv)](https:\u002F\u002Farxiv.org\u002Fabs\u002F2510.14528)\n\n[![PyPI Downloads](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FPaddlePaddle_PaddleOCR_readme_9b9d981fce5f.png)](https:\u002F\u002Fpepy.tech\u002Fprojects\u002Fpaddleocr)\n[![PyPI Downloads](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FPaddlePaddle_PaddleOCR_readme_f36ee4c7f7f6.png)](https:\u002F\u002Fpepy.tech\u002Fprojects\u002Fpaddleocr)\n[![Used by](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FUsed%20by-6k%2B%20repositories-blue)](https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fnetwork\u002Fdependents)\n[![PyPI version](https:\u002F\u002Fimg.shields.io\u002Fpypi\u002Fv\u002Fpaddleocr)](https:\u002F\u002Fpypi.org\u002Fproject\u002Fpaddleocr\u002F)\n![python](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fpython-3.8~3.13-aff.svg)\n\n![os](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fos-linux%2C%20win%2C%20mac-pink.svg)\n![hardware](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fhardware-cpu%2C%20gpu%2C%20xpu%2C%20npu-yellow.svg)\n[![License](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Flicense-Apache_2.0-green)](..\u002FLICENSE)\n[![Ask DeepWiki](https:\u002F\u002Fdeepwiki.com\u002Fbadge.svg)](https:\u002F\u002Fdeepwiki.com\u002FPaddlePaddle\u002FPaddleOCR)\n[![AI Studio](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FPaddleOCR-_Offiical_Website-1927BA?logo=data:image\u002Fpng;base64,iVBORw0KGgoAAAANSUhEUgAAAgAAAAIACAMAAADDpiTIAAAABlBMVEU2P+X\u002F\u002F\u002F+1KuUwAAAHKklEQVR42u3dS5bjOAwEwALvf2fMavZum6IAImI7b2yYSqU+1Zb\u002F\u002FgAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAADKCR\u002F+fzly7rD92yVg69xh8zeLwOa5w+ZvFYHtc4ft3ykB++cOm79PAp6YO2z\u002FNgl4ZO5l+9+yT4QAvLqS748VF33Ylzdvzpl72f6z53YIGJ6SZdPeNHcIwOycaADdLgCSIgAIgCOAACAAykIAEAAEAAFAABCAT+WQuQVgeBqXhXQIQAAYegowLQBpbg3gZGFyAC6vgBQAMREA2\u002FYfDPxyaDQNyTNz+3Zwn5J4ZG7PB2h0kHhi7plPCImmJwkPzO0RMa3OET0i5uGlzHFze0xcu0vE2Dq3J4U2vEPgSaHbFzPNDQAAAAAAAMBNovdw+cP\u002Fny+uaf7w\u002F+eYADy8kE+F4Offdjn6zZXhAXgiA78G4MNNsmnu1Xr7b3mbOL8T5Ja5bw\u002FA35EC2LiWpzt1y9jRugBy30fLg3NvHPvnuZcC2NsCUXA\u002FaRmA89V07Fwgt37uH8deCmBr6N44pP4UgaUATpdA7v\u002FcMbIB8okliY65\u002FSW5HhJ1ehPmM+8edwXgpbu4R88FayR32Y\u002FP7oZZbOx13\u002FZr\u002F\u002FZHx27bAPnkFoyewYlbAhD3TvBobr95gaUAtr1EdNx1lgI4OcTTuR3z6+FZMEDRcu9ZCuDgGCdyGxMa4EgBRMvcjrkM7NgBZw5c0TwAUWUhZwRXA2xaya65Xa3jO2qYZ8bu2AD5w38tG5V8aZpoGN6Tz0bOfa9bceyWAciTO0jWyO1Tc5cLwJmF\u002FJfPnXVyu3\u002FslgHIg1n79O2O5fZv+1cHV7sC2HYqmUdHysNzX3sVkMcjUK5Gc+dMs28E5bGtm0V3gloBOP9vgZv+4sYn3RUaYFMCol5uN77g6lUApc8pWs69Zn7snS9Z9Q8G0S0AUTVUUTG3A54R1KSvo\u002FdiLAv5fKzynZeN6xogC75u93+AtBTA47OlAFSv6qY\u002Fvp3DAjD8iv2ZdFYJwKynMhTK1rInPfzaxW81LnvSgFP9KxrATaCLA3DxHpbFX31ZyNm5XRZyXG5bNkAWfP0rcrsUwOgC6NIAzgBcBiqAWwPgLrAGuGBP6jr2sifdfiJ6QQM4Bbw4AK4B3129ZSFn53ZZyA\u002FGyFty27IBFMDFAXAG8PbyLQv5xULGPRl0K3h2AbwcgCZPhs+LD1zLnjS6AN4NwMU\u002FDVFh7LyhASreTbvqrxdr\u002FJ4XT4Swz4FrTS+AGJ7bNbwAYkxuWzZAVljHrJfbjb9wviYXwFO\u002FFJ8Vli4vaICsEMFyBbA3tmtsAUS0zG1c\u002Fbj4YwsZH2\u002F+Whd0+1Nb+S7IE2sfPw4RL0XmsR8Nqvz7qFngmPHF34EqjP15AAofAkosZKPC\u002FK6FVoeP02Ehi540NG6AK\u002F4pYP3cLgVwXwHkDQ1QcSGb\u002FuF4WwCmfX8u\u002F+4vgLINcMUlQIfcLgXwXAF0+BGkpQDuuJx7\u002Fhwgpu\u002F\u002FcWVuO3wxJOz\u002Fz8297vgYBwaIO3O7Kn+c194578ltywbIgu8fl+Z2lS+APvnLjnOv8hsgSqxjgwL4Ln9LAezaj98tgPzy7ZcC+GQzxrWxXQpgx370dm6\u002FH7v6jaBoso5dY1swAFlwHWvfBf5pxVa93fCtdx64+1dsgCy4joWvAfPX9VoKYMs6Zse9\u002F8Mlvv7LILlhAfKFFdsSutJXAdFkL3qlADJPrXFcXAC5KYaH586jO9mtAch9S3T0GQJ726ZWAE49kjP3rlDJuetdaL\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\u002FgN+IH8U6YveYgAAAABJRU5ErkJggg==&labelColor=white)](https:\u002F\u002Fwww.paddleocr.com)\n\n\n\n**PaddleOCR is an industry-leading, production-ready OCR and document AI engine, offering end-to-end solutions from text extraction to intelligent document understanding**\n\n\u003C\u002Fdiv>\n\n# PaddleOCR\n[![Framework](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FPaddlePaddle-3.0-orange)](https:\u002F\u002Fwww.paddlepaddle.org.cn\u002Fen)\n[![Accuracy](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FRecognition%20Accuracy-🏆-green)](#)\n[![Multi-Language](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FSupport_Languages-100+-brightgreen)](#)\n[![Handwriting](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FHandwriting-✓-success)](#)\n[![Hardware](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FHeterogeneous%20Hardware-Kunlunxin%20%7C%20Ascend_NPU-red)](#)\n\n> [!TIP]\n> PaddleOCR now provides an MCP server that supports integration with Agent applications like Claude Desktop. For details, please refer to [PaddleOCR MCP Server](https:\u002F\u002Fpaddlepaddle.github.io\u002FPaddleOCR\u002Flatest\u002Fen\u002Fversion3.x\u002Fdeployment\u002Fmcp_server.html).\n>\n> The PaddleOCR 3.0 Technical Report is now available. See details at: [PaddleOCR 3.0 Technical Report](https:\u002F\u002Farxiv.org\u002Fabs\u002F2507.05595).\n>\n> The PaddleOCR-VL Technical Report is now available. See details at [PaddleOCR-VL Technical Report](https:\u002F\u002Farxiv.org\u002Fabs\u002F2510.14528).\n>\n> The Beta version of the PaddleOCR official website is now live, offering a more convenient online experience and large-scale PDF file parsing, as well as free API and MCP services. For more details, please visit the [PaddleOCR official website](https:\u002F\u002Fwww.paddleocr.com).\n\n\n**PaddleOCR** converts documents and images into **structured, AI-friendly data** (like JSON and Markdown) with **industry-leading accuracy**—powering AI applications for everyone from indie developers and startups to large enterprises worldwide. With over **60,000 stars** and deep integration into leading projects like **MinerU, RAGFlow, pathway and cherry-studio**, PaddleOCR has become the **premier solution** for developers building intelligent document applications in the **AI era**.\n\n### PaddleOCR 3.0 Core Features\n\n[![Official Website](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FPaddleOCR--VL--1.5-_Official_Website-1927BA?logo=data:image\u002Fpng;base64,iVBORw0KGgoAAAANSUhEUgAAAgAAAAIACAMAAADDpiTIAAAABlBMVEU2P+X\u002F\u002F\u002F+1KuUwAAAHKklEQVR42u3dS5bjOAwEwALvf2fMavZum6IAImI7b2yYSqU+1Zb\u002F\u002FgAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAADKCR\u002F+fzly7rD92yVg69xh8zeLwOa5w+ZvFYHtc4ft3ykB++cOm79PAp6YO2z\u002FNgl4ZO5l+9+yT4QAvLqS748VF33Ylzdvzpl72f6z53YIGJ6SZdPeNHcIwOycaADdLgCSIgAIgCOAACAAykIAEAAEAAFAABCAT+WQuQVgeBqXhXQIQAAYegowLQBpbg3gZGFyAC6vgBQAMREA2\u002FYfDPxyaDQNyTNz+3Zwn5J4ZG7PB2h0kHhi7plPCImmJwkPzO0RMa3OET0i5uGlzHFze0xcu0vE2Dq3J4U2vEPgSaHbFzPNDQAAAAAAAMBNovdw+cP\u002Fny+uaf7w\u002F+eYADy8kE+F4Offdjn6zZXhAXgiA78G4MNNsmnu1Xr7b3mbOL8T5Ja5bw\u002FA35EC2LiWpzt1y9jRugBy30fLg3NvHPvnuZcC2NsCUXA\u002FaRmA89V07Fwgt37uH8deCmBr6N44pP4UgaUATpdA7v\u002FcMbIB8okliY65\u002FSW5HhJ1ehPmM+8edwXgpbu4R88FayR32Y\u002FP7oZZbOx13\u002FZr\u002F\u002FZHx27bAPnkFoyewYlbAhD3TvBobr95gaUAtr1EdNx1lgI4OcTTuR3z6+FZMEDRcu9ZCuDgGCdyGxMa4EgBRMvcjrkM7NgBZw5c0TwAUWUhZwRXA2xaya65Xa3jO2qYZ8bu2AD5w38tG5V8aZpoGN6Tz0bOfa9bceyWAciTO0jWyO1Tc5cLwJmF\u002FJfPnXVyu3\u002FslgHIg1n79O2O5fZv+1cHV7sC2HYqmUdHysNzX3sVkMcjUK5Gc+dMs28E5bGtm0V3gloBOP9vgZv+4sYn3RUaYFMCol5uN77g6lUApc8pWs69Zn7snS9Z9Q8G0S0AUTVUUTG3A54R1KSvo\u002FdiLAv5fKzynZeN6xogC75u93+AtBTA47OlAFSv6qY\u002Fvp3DAjD8iv2ZdFYJwKynMhTK1rInPfzaxW81LnvSgFP9KxrATaCLA3DxHpbFX31ZyNm5XRZyXG5bNkAWfP0rcrsUwOgC6NIAzgBcBiqAWwPgLrAGuGBP6jr2sifdfiJ6QQM4Bbw4AK4B3129ZSFn53ZZyA\u002FGyFty27IBFMDFAXAG8PbyLQv5xULGPRl0K3h2AbwcgCZPhs+LD1zLnjS6AN4NwMU\u002FDVFh7LyhASreTbvqrxdr\u002FJ4XT4Swz4FrTS+AGJ7bNbwAYkxuWzZAVljHrJfbjb9wviYXwFO\u002FFJ8Vli4vaICsEMFyBbA3tmtsAUS0zG1c\u002Fbj4YwsZH2\u002F+Whd0+1Nb+S7IE2sfPw4RL0XmsR8Nqvz7qFngmPHF34EqjP15AAofAkosZKPC\u002FK6FVoeP02Ehi540NG6AK\u002F4pYP3cLgVwXwHkDQ1QcSGb\u002FuF4WwCmfX8u\u002F+4vgLINcMUlQIfcLgXwXAF0+BGkpQDuuJx7\u002Fhwgpu\u002F\u002FcWVuO3wxJOz\u002Fz8297vgYBwaIO3O7Kn+c194578ltywbIgu8fl+Z2lS+APvnLjnOv8hsgSqxjgwL4Ln9LAezaj98tgPzy7ZcC+GQzxrWxXQpgx370dm6\u002FH7v6jaBoso5dY1swAFlwHWvfBf5pxVa93fCtdx64+1dsgCy4joWvAfPX9VoKYMs6Zse9\u002F8Mlvv7LILlhAfKFFdsSutJXAdFkL3qlADJPrXFcXAC5KYaH586jO9mtAch9S3T0GQJ726ZWAE49kjP3rlDJuetdaL\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\u002FgN+IH8U6YveYgAAAABJRU5ErkJggg==&labelColor=white)](https:\u002F\u002Faistudio.baidu.com\u002Fpaddleocr)\n[![HuggingFace](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FPaddleOCR--VL--1.5-_Demo_on_HuggingFace-yellow?logo=data:image\u002Fpng;base64,iVBORw0KGgoAAAANSUhEUgAAAF8AAABYCAMAAACkl9t\u002FAAAAk1BMVEVHcEz\u002FnQv\u002FnQv\u002FnQr\u002FnQv\u002FnQr\u002FnQv\u002FnQv\u002FnQr\u002FwRf\u002FtxT\u002Fpg7\u002FyRr\u002FrBD\u002FzRz\u002Fngv\u002FoAz\u002Fzhz\u002Fnwv\u002FtxT\u002Fngv\u002F0B3+zBz\u002FnQv\u002F0h7\u002Fwxn\u002FvRb\u002FthXkuiT\u002FrxH\u002FpxD\u002Fogzcqyf\u002FnQvTlSz\u002FczCxky7\u002FSjifdjT\u002FMj3+Mj3wMj15aTnDNz+DSD9RTUBsP0FRO0Q6O0WyIxEIAAAAGHRSTlMADB8zSWF3krDDw8TJ1NbX5efv8ff9\u002FfxKDJ9uAAAGKklEQVR42u2Z63qjOAyGC4RwCOfB2JAGqrSb2WnTw\u002F1f3UaWcSGYNKTdf\u002FP+mOkTrE+yJBulvfvLT2A5ruenaVHyIks33npl\u002F6C4s\u002FZLAM45SOi\u002F1FtZPyFur1OYofBX3w7d54Bxm+E8db+nDr12ttmESZ4zludJEG5S7TO72YPlKZFyE+YCYUJTBZsMiNS5Sd7NlDmKM2Eg2JQg8awbglfqgbhArjxkS7dgp2RH6hc9AMLdZYUtZN5DJr4molC8BfKrEkPKEnEVjLbgW1fLy77ZVOJagoIcLIl+IxaQZGjiX597HopF5CkaXVMDO9Pyix3AFV3kw4lQLCbHuMovz8FallbcQIJ5Ta0vks9RnolbCK84BtjKRS5uA43hYoZcOBGIG2Epbv6CvFVQ8m8loh66WNySsnN7htL58LNp+NXT8\u002FPhXiBXPMjLSxtwp8W9f\u002F1AngRierBkA+kk\u002FIpUSOeKByzn8y3kAAAfh\u002F\u002F0oXgV4roHm\u002Fkz4E2z\u002F\u002FzRc3\u002FlgwBzbM2mJxQEa5pqgX7d1L0htrhx7LKxOZlKbwcAWyEOWqYSI8YPtgDQVjpB5nvaHaSnBaQSD6hweDi8PosxD6\u002FPT09YY3xQA7LTCTKfYX+QHpA0GCcqmEHvr\u002FcyfKQTEuwgbs2kPxJEB0iNjfJcCTPyocx+A0griHSmADiC91oNGVwJ69RudYe65vJmoqfpul0lrqXadW0jFKH5BKwAeCq+Den7s+3zfRJzA61\u002FUj\u002F9H\u002FVzLKTx9jFPPdXeeP+L7WEvDLAKAIoF8bPTKT0+TM7W8ePj3Rz\u002FYn3kOAp2f1Kf0Weony7pn\u002FcPydvhQYV+eFOfmOu7VB\u002FViPe34\u002FEN3RFHY\u002FyRuT8ddCtMPH\u002FMcBAT5s+vRde\u002Fgf2c\u002FsPsjLK+m5IBQF5tO+h2tTlBGnP6693JdsvofjOPnnEHkh2TnV\u002FX1fBl9S5zrwuwF8NFrAVJVwCAPTe8gaJlomqlp0pv4Pjn98tJ\u002Ft\u002FfL++6unpR1YGC2n\u002FKCoa0tTLoKiEeUPDl94nj+5\u002FTv3\u002FeT5vBQ60X1S0oZr+IWRR8Ldhu7AlLjPISlJcO9vrFotky9SpzDequlwEir5beYAc0R7D9KS1DXva0jhYRDXoExPdc6yw5GShkZXe9QdO\u002FuOvHofxjrV\u002FTNS6iMJS+4TcSTgk9n5agJdBQbB\u002F\u002FIfF\u002FHpvPt3Tbi7b6I6K0R72p6ajryEJrENW2bbeVUGjfgoals4L443c7BEE4mJO2SpbRngxQrAKRudRzGQ8jVOL2qDVjjI8K1gc3TIJ5KiFZ1q+gdsARPB4NQS4AjwVSt72DSoXNyOWUrU5mQ9nRYyjp89Xo7oRI6Bga9QNT1mQ\u002FptaJq5T\u002F7WcgAZywR\u002FXlPGAUDdet3LE+qS0TI+g+aJU8MIqjo0Kx8Ly+maxLjJmjQ18rA0YCkxLQbUZP1WqdmyQGJLUm7VnQFqodmXSqmRrdVpqdzk5LvmvgtEcW8PMGdaS23EOWyDVbACZzUJPaqMbjDxpA3Qrgl0AikimGDbqmyT8P8NOYiqrldF8rX+YN7TopX4UoHuSCYY7cgX4gHwclQKl1zhx0THf+tCAUValzjI7Wg9EhptrkIcfIJjA94evOn8B2eHaVzvBrnl2ig0So6hvPaz0IGcOvTHvUIlE2+prqAxLSQxZlU2stql1NqCCLdIiIN\u002Fi1DBEHUoElM9dBravbiAnKqgpi4IBkw+utSPIoBijDXJipSVV7MpOEJUAc5Qmm3BnUN+w3hteEieYKfRZSIUcXKMVf0u5wD4EwsUNVvZOtUT7A2GkffHjByWpHqvRBYrTV72a6j8zZ6W0DTE86Hn04bmyWX3Ri9WH7ZU6Q7h+ZHo0nHUAcsQvVhXRDZHChwiyi\u002FhnPuOsSEF6Exk3o6Y9DT1eZ+6cASXk2Y9k+6EOQMDGm6WBK10wOQJCBwren86cPPWUcRAnTVjGcU1LBgs9FURiX\u002Fe6479yZcLwCBmTxiawEwrOcleuu12t3tbLv\u002FN4RLYIBhYexm7Fcn4OJcn0+zc+s8\u002FVfPeddZHAGN6TT8eGczHdR\u002FGts1\u002FMzDkThr23zqrVfAMFT33Nx1RJsx1k5zuWILLnG\u002FvsH+Fv5D4NTVcp1Gzo8AAAAAElFTkSuQmCC&labelColor=white)](https:\u002F\u002Fhuggingface.co\u002Fspaces\u002FPaddlePaddle\u002FPaddleOCR-VL-1.5_Online_Demo)\n[![ModelScope](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FPaddleOCR--VL--1.5-_Demo_on_ModelScope-purple?logo=data:image\u002Fsvg+xml;base64,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&labelColor=white)](https:\u002F\u002Fwww.modelscope.cn\u002Fstudios\u002FPaddlePaddle\u002FPaddleOCR-VL-1.5_Online_Demo)\n\n[![AI Studio](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FPP--OCRv5-Demo_on_AI_Studio-1927BA?logo=data:image\u002Fpng;base64,iVBORw0KGgoAAAANSUhEUgAAAgAAAAIACAMAAADDpiTIAAAABlBMVEU2P+X\u002F\u002F\u002F+1KuUwAAAHKklEQVR42u3dS5bjOAwEwALvf2fMavZum6IAImI7b2yYSqU+1Zb\u002F\u002FgAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAADKCR\u002F+fzly7rD92yVg69xh8zeLwOa5w+ZvFYHtc4ft3ykB++cOm79PAp6YO2z\u002FNgl4ZO5l+9+yT4QAvLqS748VF33Ylzdvzpl72f6z53YIGJ6SZdPeNHcIwOycaADdLgCSIgAIgCOAACAAykIAEAAEAAFAABCAT+WQuQVgeBqXhXQIQAAYegowLQBpbg3gZGFyAC6vgBQAMREA2\u002FYfDPxyaDQNyTNz+3Zwn5J4ZG7PB2h0kHhi7plPCImmJwkPzO0RMa3OET0i5uGlzHFze0xcu0vE2Dq3J4U2vEPgSaHbFzPNDQAAAAAAAMBNovdw+cP\u002Fny+uaf7w\u002F+eYADy8kE+F4Offdjn6zZXhAXgiA78G4MNNsmnu1Xr7b3mbOL8T5Ja5bw\u002FA35EC2LiWpzt1y9jRugBy30fLg3NvHPvnuZcC2NsCUXA\u002FaRmA89V07Fwgt37uH8deCmBr6N44pP4UgaUATpdA7v\u002FcMbIB8okliY65\u002FSW5HhJ1ehPmM+8edwXgpbu4R88FayR32Y\u002FP7oZZbOx13\u002FZr\u002F\u002FZHx27bAPnkFoyewYlbAhD3TvBobr95gaUAtr1EdNx1lgI4OcTTuR3z6+FZMEDRcu9ZCuDgGCdyGxMa4EgBRMvcjrkM7NgBZw5c0TwAUWUhZwRXA2xaya65Xa3jO2qYZ8bu2AD5w38tG5V8aZpoGN6Tz0bOfa9bceyWAciTO0jWyO1Tc5cLwJmF\u002FJfPnXVyu3\u002FslgHIg1n79O2O5fZv+1cHV7sC2HYqmUdHysNzX3sVkMcjUK5Gc+dMs28E5bGtm0V3gloBOP9vgZv+4sYn3RUaYFMCol5uN77g6lUApc8pWs69Zn7snS9Z9Q8G0S0AUTVUUTG3A54R1KSvo\u002FdiLAv5fKzynZeN6xogC75u93+AtBTA47OlAFSv6qY\u002Fvp3DAjD8iv2ZdFYJwKynMhTK1rInPfzaxW81LnvSgFP9KxrATaCLA3DxHpbFX31ZyNm5XRZyXG5bNkAWfP0rcrsUwOgC6NIAzgBcBiqAWwPgLrAGuGBP6jr2sifdfiJ6QQM4Bbw4AK4B3129ZSFn53ZZyA\u002FGyFty27IBFMDFAXAG8PbyLQv5xULGPRl0K3h2AbwcgCZPhs+LD1zLnjS6AN4NwMU\u002FDVFh7LyhASreTbvqrxdr\u002FJ4XT4Swz4FrTS+AGJ7bNbwAYkxuWzZAVljHrJfbjb9wviYXwFO\u002FFJ8Vli4vaICsEMFyBbA3tmtsAUS0zG1c\u002Fbj4YwsZH2\u002F+Whd0+1Nb+S7IE2sfPw4RL0XmsR8Nqvz7qFngmPHF34EqjP15AAofAkosZKPC\u002FK6FVoeP02Ehi540NG6AK\u002F4pYP3cLgVwXwHkDQ1QcSGb\u002FuF4WwCmfX8u\u002F+4vgLINcMUlQIfcLgXwXAF0+BGkpQDuuJx7\u002Fhwgpu\u002F\u002FcWVuO3wxJOz\u002Fz8297vgYBwaIO3O7Kn+c194578ltywbIgu8fl+Z2lS+APvnLjnOv8hsgSqxjgwL4Ln9LAezaj98tgPzy7ZcC+GQzxrWxXQpgx370dm6\u002FH7v6jaBoso5dY1swAFlwHWvfBf5pxVa93fCtdx64+1dsgCy4joWvAfPX9VoKYMs6Zse9\u002F8Mlvv7LILlhAfKFFdsSutJXAdFkL3qlADJPrXFcXAC5KYaH586jO9mtAch9S3T0GQJ726ZWAE49kjP3rlDJuetdaL\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\u002FgN+IH8U6YveYgAAAABJRU5ErkJggg==&labelColor=white)](https:\u002F\u002Faistudio.baidu.com\u002Fcommunity\u002Fapp\u002F91660\u002FwebUI)\n[![AI Studio](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FPP--StructureV3-Demo_on_AI_Studio-1927BA?logo=data:image\u002Fpng;base64,iVBORw0KGgoAAAANSUhEUgAAAgAAAAIACAMAAADDpiTIAAAABlBMVEU2P+X\u002F\u002F\u002F+1KuUwAAAHKklEQVR42u3dS5bjOAwEwALvf2fMavZum6IAImI7b2yYSqU+1Zb\u002F\u002FgAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAADKCR\u002F+fzly7rD92yVg69xh8zeLwOa5w+ZvFYHtc4ft3ykB++cOm79PAp6YO2z\u002FNgl4ZO5l+9+yT4QAvLqS748VF33Ylzdvzpl72f6z53YIGJ6SZdPeNHcIwOycaADdLgCSIgAIgCOAACAAykIAEAAEAAFAABCAT+WQuQVgeBqXhXQIQAAYegowLQBpbg3gZGFyAC6vgBQAMREA2\u002FYfDPxyaDQNyTNz+3Zwn5J4ZG7PB2h0kHhi7plPCImmJwkPzO0RMa3OET0i5uGlzHFze0xcu0vE2Dq3J4U2vEPgSaHbFzPNDQAAAAAAAMBNovdw+cP\u002Fny+uaf7w\u002F+eYADy8kE+F4Offdjn6zZXhAXgiA78G4MNNsmnu1Xr7b3mbOL8T5Ja5bw\u002FA35EC2LiWpzt1y9jRugBy30fLg3NvHPvnuZcC2NsCUXA\u002FaRmA89V07Fwgt37uH8deCmBr6N44pP4UgaUATpdA7v\u002FcMbIB8okliY65\u002FSW5HhJ1ehPmM+8edwXgpbu4R88FayR32Y\u002FP7oZZbOx13\u002FZr\u002F\u002FZHx27bAPnkFoyewYlbAhD3TvBobr95gaUAtr1EdNx1lgI4OcTTuR3z6+FZMEDRcu9ZCuDgGCdyGxMa4EgBRMvcjrkM7NgBZw5c0TwAUWUhZwRXA2xaya65Xa3jO2qYZ8bu2AD5w38tG5V8aZpoGN6Tz0bOfa9bceyWAciTO0jWyO1Tc5cLwJmF\u002FJfPnXVyu3\u002FslgHIg1n79O2O5fZv+1cHV7sC2HYqmUdHysNzX3sVkMcjUK5Gc+dMs28E5bGtm0V3gloBOP9vgZv+4sYn3RUaYFMCol5uN77g6lUApc8pWs69Zn7snS9Z9Q8G0S0AUTVUUTG3A54R1KSvo\u002FdiLAv5fKzynZeN6xogC75u93+AtBTA47OlAFSv6qY\u002Fvp3DAjD8iv2ZdFYJwKynMhTK1rInPfzaxW81LnvSgFP9KxrATaCLA3DxHpbFX31ZyNm5XRZyXG5bNkAWfP0rcrsUwOgC6NIAzgBcBiqAWwPgLrAGuGBP6jr2sifdfiJ6QQM4Bbw4AK4B3129ZSFn53ZZyA\u002FGyFty27IBFMDFAXAG8PbyLQv5xULGPRl0K3h2AbwcgCZPhs+LD1zLnjS6AN4NwMU\u002FDVFh7LyhASreTbvqrxdr\u002FJ4XT4Swz4FrTS+AGJ7bNbwAYkxuWzZAVljHrJfbjb9wviYXwFO\u002FFJ8Vli4vaICsEMFyBbA3tmtsAUS0zG1c\u002Fbj4YwsZH2\u002F+Whd0+1Nb+S7IE2sfPw4RL0XmsR8Nqvz7qFngmPHF34EqjP15AAofAkosZKPC\u002FK6FVoeP02Ehi540NG6AK\u002F4pYP3cLgVwXwHkDQ1QcSGb\u002FuF4WwCmfX8u\u002F+4vgLINcMUlQIfcLgXwXAF0+BGkpQDuuJx7\u002Fhwgpu\u002F\u002FcWVuO3wxJOz\u002Fz8297vgYBwaIO3O7Kn+c194578ltywbIgu8fl+Z2lS+APvnLjnOv8hsgSqxjgwL4Ln9LAezaj98tgPzy7ZcC+GQzxrWxXQpgx370dm6\u002FH7v6jaBoso5dY1swAFlwHWvfBf5pxVa93fCtdx64+1dsgCy4joWvAfPX9VoKYMs6Zse9\u002F8Mlvv7LILlhAfKFFdsSutJXAdFkL3qlADJPrXFcXAC5KYaH586jO9mtAch9S3T0GQJ726ZWAE49kjP3rlDJuetdaL\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\u002FgN+IH8U6YveYgAAAABJRU5ErkJggg==&labelColor=white)](https:\u002F\u002Faistudio.baidu.com\u002Fcommunity\u002Fapp\u002F518494\u002FwebUI)\n[![AI Studio](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FPP--ChatOCRv4-Demo_on_AI_Studio-1927BA?logo=data:image\u002Fpng;base64,iVBORw0KGgoAAAANSUhEUgAAAgAAAAIACAMAAADDpiTIAAAABlBMVEU2P+X\u002F\u002F\u002F+1KuUwAAAHKklEQVR42u3dS5bjOAwEwALvf2fMavZum6IAImI7b2yYSqU+1Zb\u002F\u002FgAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAADKCR\u002F+fzly7rD92yVg69xh8zeLwOa5w+ZvFYHtc4ft3ykB++cOm79PAp6YO2z\u002FNgl4ZO5l+9+yT4QAvLqS748VF33Ylzdvzpl72f6z53YIGJ6SZdPeNHcIwOycaADdLgCSIgAIgCOAACAAykIAEAAEAAFAABCAT+WQuQVgeBqXhXQIQAAYegowLQBpbg3gZGFyAC6vgBQAMREA2\u002FYfDPxyaDQNyTNz+3Zwn5J4ZG7PB2h0kHhi7plPCImmJwkPzO0RMa3OET0i5uGlzHFze0xcu0vE2Dq3J4U2vEPgSaHbFzPNDQAAAAAAAMBNovdw+cP\u002Fny+uaf7w\u002F+eYADy8kE+F4Offdjn6zZXhAXgiA78G4MNNsmnu1Xr7b3mbOL8T5Ja5bw\u002FA35EC2LiWpzt1y9jRugBy30fLg3NvHPvnuZcC2NsCUXA\u002FaRmA89V07Fwgt37uH8deCmBr6N44pP4UgaUATpdA7v\u002FcMbIB8okliY65\u002FSW5HhJ1ehPmM+8edwXgpbu4R88FayR32Y\u002FP7oZZbOx13\u002FZr\u002F\u002FZHx27bAPnkFoyewYlbAhD3TvBobr95gaUAtr1EdNx1lgI4OcTTuR3z6+FZMEDRcu9ZCuDgGCdyGxMa4EgBRMvcjrkM7NgBZw5c0TwAUWUhZwRXA2xaya65Xa3jO2qYZ8bu2AD5w38tG5V8aZpoGN6Tz0bOfa9bceyWAciTO0jWyO1Tc5cLwJmF\u002FJfPnXVyu3\u002FslgHIg1n79O2O5fZv+1cHV7sC2HYqmUdHysNzX3sVkMcjUK5Gc+dMs28E5bGtm0V3gloBOP9vgZv+4sYn3RUaYFMCol5uN77g6lUApc8pWs69Zn7snS9Z9Q8G0S0AUTVUUTG3A54R1KSvo\u002FdiLAv5fKzynZeN6xogC75u93+AtBTA47OlAFSv6qY\u002Fvp3DAjD8iv2ZdFYJwKynMhTK1rInPfzaxW81LnvSgFP9KxrATaCLA3DxHpbFX31ZyNm5XRZyXG5bNkAWfP0rcrsUwOgC6NIAzgBcBiqAWwPgLrAGuGBP6jr2sifdfiJ6QQM4Bbw4AK4B3129ZSFn53ZZyA\u002FGyFty27IBFMDFAXAG8PbyLQv5xULGPRl0K3h2AbwcgCZPhs+LD1zLnjS6AN4NwMU\u002FDVFh7LyhASreTbvqrxdr\u002FJ4XT4Swz4FrTS+AGJ7bNbwAYkxuWzZAVljHrJfbjb9wviYXwFO\u002FFJ8Vli4vaICsEMFyBbA3tmtsAUS0zG1c\u002Fbj4YwsZH2\u002F+Whd0+1Nb+S7IE2sfPw4RL0XmsR8Nqvz7qFngmPHF34EqjP15AAofAkosZKPC\u002FK6FVoeP02Ehi540NG6AK\u002F4pYP3cLgVwXwHkDQ1QcSGb\u002FuF4WwCmfX8u\u002F+4vgLINcMUlQIfcLgXwXAF0+BGkpQDuuJx7\u002Fhwgpu\u002F\u002FcWVuO3wxJOz\u002Fz8297vgYBwaIO3O7Kn+c194578ltywbIgu8fl+Z2lS+APvnLjnOv8hsgSqxjgwL4Ln9LAezaj98tgPzy7ZcC+GQzxrWxXQpgx370dm6\u002FH7v6jaBoso5dY1swAFlwHWvfBf5pxVa93fCtdx64+1dsgCy4joWvAfPX9VoKYMs6Zse9\u002F8Mlvv7LILlhAfKFFdsSutJXAdFkL3qlADJPrXFcXAC5KYaH586jO9mtAch9S3T0GQJ726ZWAE49kjP3rlDJuetdaL\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\u002FgN+IH8U6YveYgAAAABJRU5ErkJggg==&labelColor=white)](https:\u002F\u002Faistudio.baidu.com\u002Fcommunity\u002Fapp\u002F518493\u002FwebUI)\n\n- **PaddleOCR-VL-1.5: 0.9B VLM for Real-World Document Parsing and Text Spotting**\n  A SOTA and resource-efficient model designed for real-world document parsing and text spotting tasks. It achieves comprehensive leadership across six major scenarios: normal, skew, warping, scanning, varied lighting, and screen photography in document parsing task. It introduces the leading new capabilities for text spotting and seal recognition, strengthens the parsing of complex elements (such as text, tables, formulas, and charts), and expands language support to 111 languages—all while maintaining extremely low resource consumption.\n\n- **PaddleOCR-VL - Multilingual Document Parsing via a 0.9B VLM**  \n  **The SOTA and resource-efficient model tailored for document parsing**, that supports 109 languages and excels in recognizing complex elements (e.g., text, tables, formulas, and charts), while maintaining minimal resource consumption.\n\n- **PP-OCRv5 — Universal Scene Text Recognition**  \n  **Single model supports five text types** (Simplified Chinese, Traditional Chinese, English, Japanese, and Pinyin) with **13% accuracy improvement**. Solves multilingual mixed document recognition challenges.\n\n- **PP-StructureV3 — Complex Document Parsing**  \n  Intelligently converts complex PDFs and document images into **Markdown and JSON files that preserve original structure**. **Outperforms** numerous commercial solutions in public benchmarks. **Perfectly maintains document layout and hierarchical structure**.\n\n- **PP-ChatOCRv4 — Intelligent Information Extraction**  \n  Natively integrates ERNIE 4.5 to **precisely extract key information** from massive documents, with 15% accuracy improvement over previous generation. Makes documents \"**understand**\" your questions and provide accurate answers.\n\nIn addition to providing an outstanding model library, PaddleOCR 3.0 also offers user-friendly tools covering model training, inference, and service deployment, so developers can rapidly bring AI applications to production.\n\u003Cdiv align=\"center\">\n  \u003Cp>\n      \u003Cimg width=\"100%\" src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FPaddlePaddle_PaddleOCR_readme_603d058de9a9.jpg\" alt=\"PaddleOCR Architecture\">\n  \u003C\u002Fp>\n\u003C\u002Fdiv>\n\n**Special Note**: PaddleOCR 3.x introduces several significant interface changes. **Old code written based on PaddleOCR 2.x is likely incompatible with PaddleOCR 3.x**. Please ensure that the documentation you are reading matches the version of PaddleOCR you are using. [This document](https:\u002F\u002Fpaddlepaddle.github.io\u002FPaddleOCR\u002Flatest\u002Fen\u002Fupdate\u002Fupgrade_notes.html) explains the reasons for the upgrade and the major changes from PaddleOCR 2.x to 3.x.\n\n## 📣 Recent updates\n\n### 🔥🔥 2026.1.29: PaddleOCR 3.4.0 released, includes:\n\n- Released PaddleOCR-VL-1.5:\n    - **Model Introduction:**\n      - PaddleOCR-VL-1.5 is a new version of PaddleOCR-VL, with a heightened focus on document parsing capabilities in real-world scenarios and the expansion of new functionalities. Powered by the innovative **PP-DocLayoutV3** algorithm for irregular shape positioning, it achieves precise layout analysis across natural document scenarios involving skew, warping, scanning, varied lighting, and screen photography. The compact multi-modal model, **PaddleOCR-VL-1.5-0.9B**, maintains its original parameter count while expanding its leading capabilities in text spotting, and seal recognition. Language support has been extended to **111 languages**, and element recognition in complex scenarios has been significantly strengthened. The model is now available on [HuggingFace](https:\u002F\u002Fhuggingface.co\u002FPaddlePaddle\u002FPaddleOCR-VL-1.5). You can also experience it online or call the API via the [PaddleOCR Official Website](https:\u002F\u002Fwww.paddleocr.com).\n  - **Core Features:**\n    - **SOTA Performance in Document Parsing:** PaddleOCR-VL-1.5 achieved a high precision of **94.5%** on the OmniDocBench v1.5 benchmark, surpassing top-tier global general large models and specialized document parsing models.\n    - **SOTA Performance Across 5 Real-World Scenarios:** Introducing an innovative document parsing approach, it is the first to support irregular document layout positioning. It outperforms open-source and closed-source models across all the five real-world scenarios: skew, warping, scanning, varied lighting, and screen photography.\n    - **Capability Expansion Based on a 0.9B Compact Model:** With a parameter size of just 0.9B, PaddleOCR-VL-1.5 has expanded its scope to include text spotting, and seal recognition, setting new SOTA results across these related tasks.\n    - **Enhanced Multi-Element Recognition:** The model features improved recognition performance for specific scenarios and multi-language content, including special symbols, ancient texts, multi-language tables, underlines, and checkboxes. Language coverage has been expanded to include **Tibetan and Bengali**.\n    - **Long Document Cross-Page Parsing:** The model supports automatic merging of cross-page tables and the identification of cross-page paragraph headings, effectively solving the issue of content fragmentation in long document parsing.\n\n\u003Cdetails>\n\u003Csummary>\u003Cstrong>2025.10.16: Release of PaddleOCR 3.3.0\u003C\u002Fstrong>\u003C\u002Fsummary>\n\n- Released PaddleOCR-VL:\n    - **Model Introduction**:\n        - **PaddleOCR-VL** is a SOTA and resource-efficient model tailored for document parsing. Its core component is PaddleOCR-VL-0.9B, a compact yet powerful vision-language model (VLM) that integrates a NaViT-style dynamic resolution visual encoder with the ERNIE-4.5-0.3B language model to enable accurate element recognition. **This innovative model efficiently supports 109 languages and excels in recognizing complex elements (e.g., text, tables, formulas, and charts), while maintaining minimal resource consumption**. Through comprehensive evaluations on widely used public benchmarks and in-house benchmarks, PaddleOCR-VL achieves SOTA performance in both page-level document parsing and element-level recognition. It significantly outperforms existing solutions, exhibits strong competitiveness against top-tier VLMs, and delivers fast inference speeds. These strengths make it highly suitable for practical deployment in real-world scenarios. The model has been released on [HuggingFace](https:\u002F\u002Fhuggingface.co\u002FPaddlePaddle\u002FPaddleOCR-VL). Everyone is welcome to download and use it! More introduction information can be found in [PaddleOCR-VL](https:\u002F\u002Fwww.paddleocr.ai\u002Flatest\u002Fversion3.x\u002Falgorithm\u002FPaddleOCR-VL\u002FPaddleOCR-VL.html).\n\n    - **Core Features**:\n        - **Compact yet Powerful VLM Architecture**: We present a novel vision-language model that is specifically designed for resource-efficient inference, achieving outstanding performance in element recognition. By integrating a NaViT-style dynamic high-resolution visual encoder with the lightweight ERNIE-4.5-0.3B language model, we significantly enhance the model’s recognition capabilities and decoding efficiency. This integration maintains high accuracy while reducing computational demands, making it well-suited for efficient and practical document processing applications.\n        - **SOTA Performance on Document Parsing**: PaddleOCR-VL achieves state-of-the-art performance in both page-level document parsing and element-level recognition. It significantly outperforms existing pipeline-based solutions and exhibiting strong competitiveness against leading vision-language models (VLMs) in document parsing. Moreover, it excels in recognizing complex document elements, such as text, tables, formulas, and charts, making it suitable for a wide range of challenging content types, including handwritten text and historical documents. This makes it highly versatile and suitable for a wide range of document types and scenarios.\n        - **Multilingual Support**: PaddleOCR-VL Supports 109 languages, covering major global languages, including but not limited to Chinese, English, Japanese, Latin, and Korean, as well as languages with different scripts and structures, such as Russian (Cyrillic script), Arabic, Hindi (Devanagari script), and Thai. This broad language coverage substantially enhances the applicability of our system to multilingual and globalized document processing scenarios.\n\n- Released PP-OCRv5 Multilingual Recognition Model:\n    - Improved the accuracy and coverage of Latin script recognition; added support for Cyrillic, Arabic, Devanagari, Telugu, Tamil, and other language systems, covering recognition of 109 languages. The model has only 2M parameters, and the accuracy of some models has increased by over 40% compared to the previous generation.\n\n\u003C\u002Fdetails>\n\n\n\u003Cdetails>\n\u003Csummary>\u003Cstrong>2025.08.21: Release of PaddleOCR 3.2.0\u003C\u002Fstrong>\u003C\u002Fsummary>\n\n- **Significant Model Additions:**\n    - Introduced training, inference, and deployment for PP-OCRv5 recognition models in English, Thai, and Greek. **The PP-OCRv5 English model delivers an 11% improvement in English scenarios compared to the main PP-OCRv5 model, with the Thai and Greek recognition models achieving accuracies of 82.68% and 89.28%, respectively.**\n\n- **Deployment Capability Upgrades:**\n    - **Full support for PaddlePaddle framework versions 3.1.0 and 3.1.1.**\n    - **Comprehensive upgrade of the PP-OCRv5 C++ local deployment solution, now supporting both Linux and Windows, with feature parity and identical accuracy to the Python implementation.**\n    - **High-performance inference now supports CUDA 12, and inference can be performed using either the Paddle Inference or ONNX Runtime backends.**\n    - **The high-stability service-oriented deployment solution is now fully open-sourced, allowing users to customize Docker images and SDKs as required.**\n    - The high-stability service-oriented deployment solution also supports invocation via manually constructed HTTP requests, enabling client-side code development in any programming language.\n\n- **Benchmark Support:**\n    - **All production lines now support fine-grained benchmarking, enabling measurement of end-to-end inference time as well as per-layer and per-module latency data to assist with performance analysis. [Here's](docs\u002Fversion3.x\u002Fpipeline_usage\u002Finstructions\u002Fbenchmark.en.md) how to set up and use the benchmark feature.**\n    - **Documentation has been updated to include key metrics for commonly used configurations on mainstream hardware, such as inference latency and memory usage, providing deployment references for users.**\n\n- **Bug Fixes:**\n    - Resolved the issue of failed log saving during model training.\n    - Upgraded the data augmentation component for formula models for compatibility with newer versions of the albumentations dependency, and fixed deadlock warnings when using the tokenizers package in multi-process scenarios.\n    - Fixed inconsistencies in switch behaviors (e.g., `use_chart_parsing`) in the PP-StructureV3 configuration files compared to other pipelines.\n\n- **Other Enhancements:**\n    - **Separated core and optional dependencies. Only minimal core dependencies are required for basic text recognition; additional dependencies for document parsing and information extraction can be installed as needed.**\n    - **Enabled support for NVIDIA RTX 50 series graphics cards on Windows; users can refer to the [installation guide](docs\u002Fversion3.x\u002Finstallation.en.md) for the corresponding PaddlePaddle framework versions.**\n    - **PP-OCR series models now support returning single-character coordinates.**\n    - Added AIStudio, ModelScope, and other model download sources, allowing users to specify the source for model downloads.\n    - Added support for chart-to-table conversion via the PP-Chart2Table module.\n    - Optimized documentation descriptions to improve usability.\n\u003C\u002Fdetails>\n\n\u003Cdetails>\n\u003Csummary>\u003Cstrong>2025.08.15: PaddleOCR 3.1.1 Released\u003C\u002Fstrong>\u003C\u002Fsummary>\n\n- **Bug Fixes:**\n  - Added the missing methods `save_vector`, `save_visual_info_list`, `load_vector`, and `load_visual_info_list` in the `PP-ChatOCRv4` class.\n  - Added the missing parameters `glossary` and `llm_request_interval` to the `translate` method in the `PPDocTranslation` class.\n\n- **Documentation Improvements:**\n  - Added a demo to the MCP documentation.\n  - Added information about the PaddlePaddle and PaddleOCR version used for performance metrics testing in the documentation.\n  - Fixed errors and omissions in the production line document translation.\n\n- **Others:**\n  - Changed the MCP server dependency to use the pure Python library `puremagic` instead of `python-magic` to reduce installation issues.\n  - Retested PP-OCRv5 performance metrics with PaddleOCR version 3.1.0 and updated the documentation.\n\n\u003C\u002Fdetails>\n\n\u003Cdetails>\n\u003Csummary>\u003Cstrong>2025.06.29: PaddleOCR 3.1.0 Released\u003C\u002Fstrong>\u003C\u002Fsummary>\n\n- **Key Models and Pipelines:**\n  - **Added PP-OCRv5 Multilingual Text Recognition Model**, which supports the training and inference process for text recognition models in 37 languages, including French, Spanish, Portuguese, Russian, Korean, etc. **Average accuracy improved by over 30%.** [Details](https:\u002F\u002Fpaddlepaddle.github.io\u002FPaddleOCR\u002Flatest\u002Fen\u002Fversion3.x\u002Falgorithm\u002FPP-OCRv5\u002FPP-OCRv5_multi_languages.html)\n  - Upgraded the **PP-Chart2Table model** in PP-StructureV3, further enhancing the capability of converting charts to tables. On internal custom evaluation sets, the metric (RMS-F1) **increased by 9.36 percentage points (71.24% -> 80.60%).**\n  - Newly launched **document translation pipeline, PP-DocTranslation, based on PP-StructureV3 and ERNIE 4.5**, which supports the translation of Markdown format documents, various complex-layout PDF documents, and document images, with the results saved as Markdown format documents. [Details](https:\u002F\u002Fpaddlepaddle.github.io\u002FPaddleOCR\u002Flatest\u002Fen\u002Fversion3.x\u002Fpipeline_usage\u002FPP-DocTranslation.html)\n\n\n- **New MCP server:** [Details](https:\u002F\u002Fpaddlepaddle.github.io\u002FPaddleOCR\u002Flatest\u002Fen\u002Fversion3.x\u002Fdeployment\u002Fmcp_server.html)\n  - **Supports both OCR and PP-StructureV3 pipelines.**\n  - Supports three working modes: local Python library, AIStudio Community Cloud Service, and self-hosted service.\n  - Supports invoking local services via stdio and remote services via Streamable HTTP.\n\n- **Documentation Optimization:** Improved the descriptions in some user guides for a smoother reading experience.\n\n\u003C\u002Fdetails>\n\n\u003Cdetails>\n    \u003Csummary>\u003Cstrong>2025.06.26: PaddleOCR 3.0.3 Released\u003C\u002Fstrong>\u003C\u002Fsummary>\n- Bug Fix: Resolved the issue where the `enable_mkldnn` parameter was not effective, restoring the default behavior of using MKL-DNN for CPU inference.\n\u003C\u002Fdetails>\n\n\u003Cdetails>\n    \u003Csummary>\u003Cstrong>2025.06.19: PaddleOCR 3.0.2 Released\u003C\u002Fstrong>\u003C\u002Fsummary>\n- **New Features:**\n\n  - The default download source has been changed from `BOS` to `HuggingFace`. Users can also change the environment variable `PADDLE_PDX_MODEL_SOURCE` to `BOS` to set the model download source back to Baidu Object Storage (BOS).\n  - Added service invocation examples for six languages—C++, Java, Go, C#, Node.js, and PHP—for pipelines like PP-OCRv5, PP-StructureV3, and PP-ChatOCRv4.\n  - Improved the layout partition sorting algorithm in the PP-StructureV3 pipeline, enhancing the sorting logic for complex vertical layouts to deliver better results.\n  - Enhanced model selection logic: when a language is specified but a model version is not, the system will automatically select the latest model version supporting that language. \n  - Set a default upper limit for MKL-DNN cache size to prevent unlimited growth, while also allowing users to configure cache capacity.\n  - Updated default configurations for high-performance inference to support Paddle MKL-DNN acceleration and optimized the logic for automatic configuration selection for smarter choices.\n  - Adjusted the logic for obtaining the default device to consider the actual support for computing devices by the installed Paddle framework, making program behavior more intuitive.\n  - Added Android example for PP-OCRv5. [Details](https:\u002F\u002Fpaddlepaddle.github.io\u002FPaddleOCR\u002Flatest\u002Fen\u002Fversion3.x\u002Fdeployment\u002Fon_device_deployment.html).\n\n- **Bug Fixes:**\n  - Fixed an issue with some CLI parameters in PP-StructureV3 not taking effect.\n  - Resolved an issue where `export_paddlex_config_to_yaml` would not function correctly in certain cases.\n  - Corrected the discrepancy between the actual behavior of `save_path` and its documentation description.\n  - Fixed potential multithreading errors when using MKL-DNN in basic service deployment.\n  - Corrected channel order errors in image preprocessing for the Latex-OCR model.\n  - Fixed channel order errors in saving visualized images within the text recognition module.\n  - Resolved channel order errors in visualized table results within PP-StructureV3 pipeline.\n  - Fixed an overflow issue in the calculation of `overlap_ratio` under extremely special circumstances in the PP-StructureV3 pipeline.\n\n- **Documentation Improvements:**\n  - Updated the description of the `enable_mkldnn` parameter in the documentation to accurately reflect the program's actual behavior.\n  - Fixed errors in the documentation regarding the `lang` and `ocr_version` parameters.\n  - Added instructions for exporting pipeline configuration files via CLI.\n  - Fixed missing columns in the performance data table for PP-OCRv5.\n  - Refined benchmark metrics for PP-StructureV3 across different configurations.\n\n- **Others:**\n\n  - Relaxed version restrictions on dependencies like numpy and pandas, restoring support for Python 3.12.\n\u003C\u002Fdetails>\n\n\u003Cdetails>\n    \u003Csummary>\u003Cstrong>History Log\u003C\u002Fstrong>\u003C\u002Fsummary>\n\n2025.06.05: **PaddleOCR 3.0.1 Released**, includes:\n\n- **Optimisation of certain models and model configurations:**\n  - Updated the default model configuration for PP-OCRv5, changing both detection and recognition from mobile to server models. To improve default performance in most scenarios, the parameter `limit_side_len` in the configuration has been changed from 736 to 64.\n  - Added a new text line orientation classification model `PP-LCNet_x1_0_textline_ori` with an accuracy of 99.42%. The default text line orientation classifier for OCR, PP-StructureV3, and PP-ChatOCRv4 pipelines has been updated to this model.\n  - Optimized the text line orientation classification model `PP-LCNet_x0_25_textline_ori`, improving accuracy by 3.3 percentage points to a current accuracy of 98.85%.\n- **Optimizations and fixes for some issues in version 3.0.0, [details](https:\u002F\u002Fpaddlepaddle.github.io\u002FPaddleOCR\u002Flatest\u002Fen\u002Fupdate\u002Fupdate.html)**\n\n🔥🔥2025.05.20: Official Release of **PaddleOCR v3.0**, including:\n- **PP-OCRv5**: High-Accuracy Text Recognition Model for All Scenarios - Instant Text from Images\u002FPDFs.\n   1. 🌐 Single-model support for **five** text types - Seamlessly process **Simplified Chinese, Traditional Chinese, Simplified Chinese Pinyin, English** and **Japanese** within a single model.\n   2. ✍️ Improved **handwriting recognition**: Significantly better at complex cursive scripts and non-standard handwriting.\n   3. 🎯 **13-point accuracy gain** over PP-OCRv4, achieving state-of-the-art performance across a variety of real-world scenarios.\n\n- **PP-StructureV3**: General-Purpose Document Parsing – Unleash SOTA Images\u002FPDFs Parsing for Real-World Scenarios! \n   1. 🧮 **High-Accuracy multi-scene PDF parsing**, leading both open- and closed-source solutions on the OmniDocBench benchmark.\n   2. 🧠 Specialized capabilities include **seal recognition**, **chart-to-table conversion**, **table recognition with nested formulas\u002Fimages**, **vertical text document parsing**, and **complex table structure analysis**.\n\n- **PP-ChatOCRv4**: Intelligent Document Understanding – Extract Key Information, not just text from Images\u002FPDFs.\n   1. 🔥 **15-point accuracy gain** in key-information extraction on PDF\u002FPNG\u002FJPG files over the previous generation.\n   2. 💻 Native support for **ERNIE 4.5**, with compatibility for large-model deployments via PaddleNLP, Ollama, vLLM, and more.\n   3. 🤝 Integrated [PP-DocBee2](https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleMIX\u002Ftree\u002Fdevelop\u002Fpaddlemix\u002Fexamples\u002Fppdocbee2), enabling extraction and understanding of printed text, handwriting, seals, tables, charts, and other common elements in complex documents.\n\n[History Log](https:\u002F\u002Fpaddlepaddle.github.io\u002FPaddleOCR\u002Flatest\u002Fen\u002Fupdate\u002Fupdate.html)\n\n\u003C\u002Fdetails>\n\n## ⚡ Quick Start\n### 1. Run online demo \n[![AI Studio](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FPP_OCRv5-AI_Studio-green)](https:\u002F\u002Faistudio.baidu.com\u002Fcommunity\u002Fapp\u002F91660\u002FwebUI)\n[![AI Studio](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FPP_StructureV3-AI_Studio-green)](https:\u002F\u002Faistudio.baidu.com\u002Fcommunity\u002Fapp\u002F518494\u002FwebUI)\n[![AI Studio](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FPP_ChatOCRv4-AI_Studio-green)](https:\u002F\u002Faistudio.baidu.com\u002Fcommunity\u002Fapp\u002F518493\u002FwebUI)\n\n### 2. Installation\n\nInstall PaddlePaddle refer to [Installation Guide](https:\u002F\u002Fwww.paddlepaddle.org.cn\u002Fen\u002Finstall\u002Fquick?docurl=\u002Fdocumentation\u002Fdocs\u002Fen\u002Fdevelop\u002Finstall\u002Fpip\u002Flinux-pip_en.html), after then, install the PaddleOCR toolkit.\n\n```bash\n# If you only want to use the basic text recognition feature (returns text position coordinates and content), including the PP-OCR series\npython -m pip install paddleocr\n# If you want to use all features such as document parsing, document understanding, document translation, key information extraction, etc.\n# python -m pip install \"paddleocr[all]\"\n```\n\nStarting from version 3.2.0, in addition to the `all` dependency group demonstrated above, PaddleOCR also supports installing partial optional features by specifying other dependency groups. All dependency groups provided by PaddleOCR are as follows:\n\n| Dependency Group Name | Corresponding Functionality |\n| - | - |\n| `doc-parser` | Document parsing: can be used to extract layout elements such as tables, formulas, stamps, images, etc. from documents; includes models like PP-StructureV3, PaddleOCR-VL |\n| `ie` | Information extraction: can be used to extract key information from documents, such as names, dates, addresses, amounts, etc.; includes models like PP-ChatOCRv4 |\n| `trans` | Document translation: can be used to translate documents from one language to another; includes models like PP-DocTranslation |\n| `all` | Complete functionality |\n\n### 3. Run inference by CLI\n```bash\n# Run PP-OCRv5 inference\npaddleocr ocr -i https:\u002F\u002Fpaddle-model-ecology.bj.bcebos.com\u002Fpaddlex\u002Fimgs\u002Fdemo_image\u002Fgeneral_ocr_002.png --use_doc_orientation_classify False --use_doc_unwarping False --use_textline_orientation False  \n\n# Run PP-StructureV3 inference\npaddleocr pp_structurev3 -i https:\u002F\u002Fpaddle-model-ecology.bj.bcebos.com\u002Fpaddlex\u002Fimgs\u002Fdemo_image\u002Fpp_structure_v3_demo.png --use_doc_orientation_classify False --use_doc_unwarping False\n\n# Get the Qianfan API Key at first, and then run PP-ChatOCRv4 inference\npaddleocr pp_chatocrv4_doc -i https:\u002F\u002Fpaddle-model-ecology.bj.bcebos.com\u002Fpaddlex\u002Fimgs\u002Fdemo_image\u002Fvehicle_certificate-1.png -k 驾驶室准乘人数 --qianfan_api_key your_api_key --use_doc_orientation_classify False --use_doc_unwarping False \n\n# Run PaddleOCR-VL inference\npaddleocr doc_parser -i https:\u002F\u002Fpaddle-model-ecology.bj.bcebos.com\u002Fpaddlex\u002Fimgs\u002Fdemo_image\u002Fpaddleocr_vl_demo.png\n\n# Get more information about \"paddleocr ocr\"\npaddleocr ocr --help\n```\n\n### 4. Run inference by API\n**4.1 PP-OCRv5 Example**\n```python\n# Initialize PaddleOCR instance\nfrom paddleocr import PaddleOCR\nocr = PaddleOCR(\n    use_doc_orientation_classify=False,\n    use_doc_unwarping=False,\n    use_textline_orientation=False)\n\n# Run OCR inference on a sample image \nresult = ocr.predict(\n    input=\"https:\u002F\u002Fpaddle-model-ecology.bj.bcebos.com\u002Fpaddlex\u002Fimgs\u002Fdemo_image\u002Fgeneral_ocr_002.png\")\n\n# Visualize the results and save the JSON results\nfor res in result:\n    res.print()\n    res.save_to_img(\"output\")\n    res.save_to_json(\"output\")\n```\n\n\u003Cdetails>\n    \u003Csummary>\u003Cstrong>4.2 PP-StructureV3 Example\u003C\u002Fstrong>\u003C\u002Fsummary>\n\n```python\nfrom pathlib import Path\nfrom paddleocr import PPStructureV3\n\npipeline = PPStructureV3(\n    use_doc_orientation_classify=False,\n    use_doc_unwarping=False\n)\n\n# For Image\noutput = pipeline.predict(\n    input=\"https:\u002F\u002Fpaddle-model-ecology.bj.bcebos.com\u002Fpaddlex\u002Fimgs\u002Fdemo_image\u002Fpp_structure_v3_demo.png\",\n)\n\n# Visualize the results and save the JSON results\nfor res in output:\n    res.print() \n    res.save_to_json(save_path=\"output\") \n    res.save_to_markdown(save_path=\"output\")           \n```\n\n\u003C\u002Fdetails>\n\n\u003Cdetails>\n   \u003Csummary>\u003Cstrong>4.3 PP-ChatOCRv4 Example\u003C\u002Fstrong>\u003C\u002Fsummary>\n\n```python\nfrom paddleocr import PPChatOCRv4Doc\n\nchat_bot_config = {\n    \"module_name\": \"chat_bot\",\n    \"model_name\": \"ernie-3.5-8k\",\n    \"base_url\": \"https:\u002F\u002Fqianfan.baidubce.com\u002Fv2\",\n    \"api_type\": \"openai\",\n    \"api_key\": \"api_key\",  # your api_key\n}\n\nretriever_config = {\n    \"module_name\": \"retriever\",\n    \"model_name\": \"embedding-v1\",\n    \"base_url\": \"https:\u002F\u002Fqianfan.baidubce.com\u002Fv2\",\n    \"api_type\": \"qianfan\",\n    \"api_key\": \"api_key\",  # your api_key\n}\n\npipeline = PPChatOCRv4Doc(\n    use_doc_orientation_classify=False,\n    use_doc_unwarping=False\n)\n\nvisual_predict_res = pipeline.visual_predict(\n    input=\"https:\u002F\u002Fpaddle-model-ecology.bj.bcebos.com\u002Fpaddlex\u002Fimgs\u002Fdemo_image\u002Fvehicle_certificate-1.png\",\n    use_common_ocr=True,\n    use_seal_recognition=True,\n    use_table_recognition=True,\n)\n\nmllm_predict_info = None\nuse_mllm = False\n# If a multimodal large model is used, the local mllm service needs to be started. You can refer to the documentation: https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleX\u002Fblob\u002Frelease\u002F3.0\u002Fdocs\u002Fpipeline_usage\u002Ftutorials\u002Fvlm_pipelines\u002Fdoc_understanding.en.md performs deployment and updates the mllm_chat_bot_config configuration.\nif use_mllm:\n    mllm_chat_bot_config = {\n        \"module_name\": \"chat_bot\",\n        \"model_name\": \"PP-DocBee\",\n        \"base_url\": \"http:\u002F\u002F127.0.0.1:8080\u002F\",  # your local mllm service url\n        \"api_type\": \"openai\",\n        \"api_key\": \"api_key\",  # your api_key\n    }\n\n    mllm_predict_res = pipeline.mllm_pred(\n        input=\"https:\u002F\u002Fpaddle-model-ecology.bj.bcebos.com\u002Fpaddlex\u002Fimgs\u002Fdemo_image\u002Fvehicle_certificate-1.png\",\n        key_list=[\"驾驶室准乘人数\"],\n        mllm_chat_bot_config=mllm_chat_bot_config,\n    )\n    mllm_predict_info = mllm_predict_res[\"mllm_res\"]\n\nvisual_info_list = []\nfor res in visual_predict_res:\n    visual_info_list.append(res[\"visual_info\"])\n    layout_parsing_result = res[\"layout_parsing_result\"]\n\nvector_info = pipeline.build_vector(\n    visual_info_list, flag_save_bytes_vector=True, retriever_config=retriever_config\n)\nchat_result = pipeline.chat(\n    key_list=[\"驾驶室准乘人数\"],\n    visual_info=visual_info_list,\n    vector_info=vector_info,\n    mllm_predict_info=mllm_predict_info,\n    chat_bot_config=chat_bot_config,\n    retriever_config=retriever_config,\n)\nprint(chat_result)\n```\n\n\u003C\u002Fdetails>\n\n\u003Cdetails>\n   \u003Csummary>\u003Cstrong>4.4 PaddleOCR-VL Example\u003C\u002Fstrong>\u003C\u002Fsummary>\n\n```python\nfrom paddleocr import PaddleOCRVL\n\npipeline = PaddleOCRVL()\noutput = pipeline.predict(\"https:\u002F\u002Fpaddle-model-ecology.bj.bcebos.com\u002Fpaddlex\u002Fimgs\u002Fdemo_image\u002Fpaddleocr_vl_demo.png\")\nfor res in output:\n    res.print()\n    res.save_to_json(save_path=\"output\")\n    res.save_to_markdown(save_path=\"output\")\n```\n\n\u003C\u002Fdetails>\n\n### 5. Chinese Heterogeneous AI Accelerators\n- [Huawei Ascend](https:\u002F\u002Fpaddlepaddle.github.io\u002FPaddleOCR\u002Flatest\u002Fversion3.x\u002Fother_devices_support\u002Fpaddlepaddle_install_NPU.html)\n- [Kunlunxin](https:\u002F\u002Fpaddlepaddle.github.io\u002FPaddleOCR\u002Flatest\u002Fversion3.x\u002Fother_devices_support\u002Fpaddlepaddle_install_XPU.html)\n\n## 🧩 More Features\n\n- Convert models to ONNX format: [Obtaining ONNX Models](https:\u002F\u002Fpaddlepaddle.github.io\u002FPaddleOCR\u002Flatest\u002Fen\u002Fversion3.x\u002Fdeployment\u002Fobtaining_onnx_models.html).\n- Accelerate inference using engines like OpenVINO, ONNX Runtime, TensorRT, or perform inference using ONNX format models: [High-Performance Inference](https:\u002F\u002Fpaddlepaddle.github.io\u002FPaddleOCR\u002Flatest\u002Fen\u002Fversion3.x\u002Fdeployment\u002Fhigh_performance_inference.html).\n- Accelerate inference using multi-GPU and multi-process: [Parallel Inference for Pipelines](https:\u002F\u002Fpaddlepaddle.github.io\u002FPaddleOCR\u002Flatest\u002Fen\u002Fversion3.x\u002Fpipeline_usage\u002Finstructions\u002Fparallel_inference.html).\n- Integrate PaddleOCR into applications written in C++, C#, Java, etc.: [Serving](https:\u002F\u002Fpaddlepaddle.github.io\u002FPaddleOCR\u002Flatest\u002Fen\u002Fversion3.x\u002Fdeployment\u002Fserving.html).\n\n## ⛰️ Advanced Tutorials\n\n- [PP-OCRv5 Tutorial](https:\u002F\u002Fpaddlepaddle.github.io\u002FPaddleOCR\u002Flatest\u002Fversion3.x\u002Fpipeline_usage\u002FOCR.html)\n- [PP-StructureV3 Tutorial](https:\u002F\u002Fpaddlepaddle.github.io\u002FPaddleOCR\u002Flatest\u002Fversion3.x\u002Fpipeline_usage\u002FPP-StructureV3.html)\n- [PP-ChatOCRv4 Tutorial](https:\u002F\u002Fpaddlepaddle.github.io\u002FPaddleOCR\u002Flatest\u002Fversion3.x\u002Fpipeline_usage\u002FPP-ChatOCRv4.html)\n- [PaddleOCR-VL Tutorial](https:\u002F\u002Fpaddlepaddle.github.io\u002FPaddleOCR\u002Flatest\u002Fversion3.x\u002Fpipeline_usage\u002FPaddleOCR-VL.html)\n\n## 🔄 Quick Overview of Execution Results\n\n### PP-OCRv5\n\n\u003Cdiv align=\"center\">\n  \u003Cp>\n       \u003Cimg width=\"100%\" src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FPaddlePaddle_PaddleOCR_readme_95355b710b08.gif\" alt=\"PP-OCRv5 Demo\">\n  \u003C\u002Fp>\n\u003C\u002Fdiv>\n\n\n\n### PP-StructureV3\n\n\u003Cdiv align=\"center\">\n  \u003Cp>\n      \u003Cimg width=\"100%\" src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FPaddlePaddle_PaddleOCR_readme_446c93f0ec25.gif\" alt=\"PP-StructureV3 Demo\">\n  \u003C\u002Fp>\n\u003C\u002Fdiv>\n\n### PaddleOCR-VL\n\n\u003Cdiv align=\"center\">\n  \u003Cp>\n      \u003Cimg width=\"100%\" src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FPaddlePaddle_PaddleOCR_readme_eba600ad581d.gif\" alt=\"PP-StructureV3 Demo\">\n  \u003C\u002Fp>\n\u003C\u002Fdiv>\n\n\n## ✨ Stay Tuned\n\n⭐ **Star this repository to keep up with exciting updates and new releases, including powerful OCR and document parsing capabilities!** ⭐\n\n\u003Cdiv align=\"center\">\n  \u003Cp>\n       \u003Cimg width=\"1200\" src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FPaddlePaddle_PaddleOCR_readme_d8a33a27b69b.gif\" alt=\"Star-Project\">\n  \u003C\u002Fp>\n\u003C\u002Fdiv>\n\n## 👩‍👩‍👧‍👦 Community\n\n\u003Cdiv align=\"center\">\n\n| PaddlePaddle WeChat official account |  Join the tech discussion group |\n| :---: | :---: |\n| \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FPaddlePaddle_PaddleOCR_readme_1bc629737e28.jpg\" width=\"150\"> | \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FPaddlePaddle_PaddleOCR_readme_6e9b6bc2266e.jpg\" width=\"150\"> |\n\u003C\u002Fdiv>\n\n\n## 😃 Awesome Projects Leveraging PaddleOCR\nPaddleOCR wouldn't be where it is today without its incredible community! 💗 A massive thank you to all our longtime partners, new collaborators, and everyone who's poured their passion into PaddleOCR — whether we've named you or not. Your support fuels our fire!\n\n\u003Cdiv align=\"center\">\n\n| Project Name | Description |\n| ------------ | ----------- |\n| [RAGFlow](https:\u002F\u002Fgithub.com\u002Finfiniflow\u002Fragflow) \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Finfiniflow\u002Fragflow\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002Finfiniflow\u002Fragflow\">\u003C\u002Fa>|RAG engine based on deep document understanding.|\n| [pathway](https:\u002F\u002Fgithub.com\u002Fpathwaycom\u002Fpathway) \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fpathwaycom\u002Fpathway\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002Fpathwaycom\u002Fpathway\">\u003C\u002Fa>|Python ETL framework for stream processing, real-time analytics, LLM pipelines, and RAG.|\n| [MinerU](https:\u002F\u002Fgithub.com\u002Fopendatalab\u002FMinerU) \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fopendatalab\u002FMinerU\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002Fopendatalab\u002FMinerU\">\u003C\u002Fa>|Multi-type Document to Markdown Conversion Tool|\n| [Umi-OCR](https:\u002F\u002Fgithub.com\u002Fhiroi-sora\u002FUmi-OCR) \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fhiroi-sora\u002FUmi-OCR\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002Fhiroi-sora\u002FUmi-OCR\">\u003C\u002Fa>|Free, Open-source, Batch Offline OCR Software.|\n| [cherry-studio](https:\u002F\u002Fgithub.com\u002FCherryHQ\u002Fcherry-studio) \u003Ca href=\"https:\u002F\u002Fgithub.com\u002FCherryHQ\u002Fcherry-studio\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002FCherryHQ\u002Fcherry-studio\">\u003C\u002Fa>|A desktop client that supports for multiple LLM providers.|\n| [OmniParser](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002FOmniParser)\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002FOmniParser\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002Fmicrosoft\u002FOmniParser\">\u003C\u002Fa> |OmniParser: Screen Parsing tool for Pure Vision Based GUI Agent.|\n| [QAnything](https:\u002F\u002Fgithub.com\u002Fnetease-youdao\u002FQAnything)\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fnetease-youdao\u002FQAnything\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002Fnetease-youdao\u002FQAnything\">\u003C\u002Fa> |Question and Answer based on Anything.|\n| [PDF-Extract-Kit](https:\u002F\u002Fgithub.com\u002Fopendatalab\u002FPDF-Extract-Kit) \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fopendatalab\u002FPDF-Extract-Kit\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002Fopendatalab\u002FPDF-Extract-Kit\">\u003C\u002Fa>|A powerful open-source toolkit designed to efficiently extract high-quality content from complex and diverse PDF documents.|\n| [Dango-Translator](https:\u002F\u002Fgithub.com\u002FPantsuDango\u002FDango-Translator)\u003Ca href=\"https:\u002F\u002Fgithub.com\u002FPantsuDango\u002FDango-Translator\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002FPantsuDango\u002FDango-Translator\">\u003C\u002Fa> |Recognize text on the screen, translate it and show the translation results in real time.|\n| [Learn more projects](.\u002Fawesome_projects.md) | [More projects based on PaddleOCR](.\u002Fawesome_projects.md)|\n\u003C\u002Fdiv>\n\n## 👩‍👩‍👧‍👦 Contributors\n\n\u003Cdiv align=\"center\">\n\u003Ca href=\"https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fgraphs\u002Fcontributors\">\n  \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FPaddlePaddle_PaddleOCR_readme_4c66843d8eb5.png\"  width=\"800\"\u002F>\n\u003C\u002Fa>\n\u003C\u002Fdiv>\n\n## 🌟 Star\n\n\u003Cdiv align=\"center\">\n  \u003Cp>\n      \u003Cimg width=\"800\" src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FPaddlePaddle_PaddleOCR_readme_1606bf1e0578.png\" alt=\"Star-history\">\n  \u003C\u002Fp>\n\u003C\u002Fdiv>\n\n\n## 📄 License\nThis project is released under the [Apache 2.0 license](LICENSE).\n\n## 🎓 Citation\n\n```bibtex\n@misc{cui2025paddleocr30technicalreport,\n      title={PaddleOCR 3.0 Technical Report}, \n      author={Cheng Cui and Ting Sun and Manhui Lin and Tingquan Gao and Yubo Zhang and Jiaxuan Liu and Xueqing Wang and Zelun Zhang and Changda Zhou and Hongen Liu and Yue Zhang and Wenyu Lv and Kui Huang and Yichao Zhang and Jing Zhang and Jun Zhang and Yi Liu and Dianhai Yu and Yanjun Ma},\n      year={2025},\n      eprint={2507.05595},\n      archivePrefix={arXiv},\n      primaryClass={cs.CV},\n      url={https:\u002F\u002Farxiv.org\u002Fabs\u002F2507.05595}, \n}\n\n@misc{cui2025paddleocrvlboostingmultilingualdocument,\n      title={PaddleOCR-VL: Boosting Multilingual Document Parsing via a 0.9B Ultra-Compact Vision-Language Model}, \n      author={Cheng Cui and Ting Sun and Suyin Liang and Tingquan Gao and Zelun Zhang and Jiaxuan Liu and Xueqing Wang and Changda Zhou and Hongen Liu and Manhui Lin and Yue Zhang and Yubo Zhang and Handong Zheng and Jing Zhang and Jun Zhang and Yi Liu and Dianhai Yu and Yanjun Ma},\n      year={2025},\n      eprint={2510.14528},\n      archivePrefix={arXiv},\n      primaryClass={cs.CV},\n      url={https:\u002F\u002Farxiv.org\u002Fabs\u002F2510.14528}, \n}\n```\n","\u003Cdiv align=\"center\">\n  \u003Cp>\n      \u003Cimg width=\"100%\" src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FPaddlePaddle_PaddleOCR_readme_f1dc4ea647a1.png\" alt=\"PaddleOCR 横幅\">\n  \u003C\u002Fp>\n\n英语 | [简体中文](.\u002Freadme\u002FREADME_cn.md) | [繁体中文](.\u002Freadme\u002FREADME_tcn.md) | [日语](.\u002Freadme\u002FREADME_ja.md) | [韩语](.\u002Freadme\u002FREADME_ko.md) | [法语](.\u002Freadme\u002FREADME_fr.md) | [俄语](.\u002Freadme\u002FREADME_ru.md) | [西班牙语](.\u002Freadme\u002FREADME_es.md) | [阿拉伯语](.\u002Freadme\u002FREADME_ar.md)\n\n\u003C!-- icon -->\n[![星标](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002FPaddlePaddle\u002FPaddleOCR?color=ccf)](https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR)\n[![分支](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fforks\u002FPaddlePaddle\u002FPaddleOCR.svg)](https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR)\n[![arXiv](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FPaddleOCR_3.0-技术报告-b31b1b.svg?logo=arXiv)](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2507.05595)\n[![arXiv](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FPaddleOCR--VL-技术报告-b31b1b.svg?logo=arXiv)](https:\u002F\u002Farxiv.org\u002Fabs\u002F2510.14528)\n\n[![PyPI 下载量](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FPaddlePaddle_PaddleOCR_readme_9b9d981fce5f.png)](https:\u002F\u002Fpepy.tech\u002Fprojects\u002Fpaddleocr)\n[![PyPI 下载量](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FPaddlePaddle_PaddleOCR_readme_f36ee4c7f7f6.png)](https:\u002F\u002Fpepy.tech\u002Fprojects\u002Fpaddleocr)\n[![使用者](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FUsed%20by-6k%2B%20repositories-blue)](https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fnetwork\u002Fdependents)\n[![PyPI 版本](https:\u002F\u002Fimg.shields.io\u002Fpypi\u002Fv\u002Fpaddleocr)](https:\u002F\u002Fpypi.org\u002Fproject\u002Fpaddleocr\u002F)\n![Python](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fpython-3.8~3.13-aff.svg)\n\n![操作系统](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fos-linux%2C%20win%2C%20mac-pink.svg)\n![硬件](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fhardware-cpu%2C%20gpu%2C%20xpu%2C%20npu-yellow.svg)\n[![许可证](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Flicense-Apache_2.0-green)](..\u002FLICENSE)\n[![咨询 DeepWiki](https:\u002F\u002Fdeepwiki.com\u002Fbadge.svg)](https:\u002F\u002Fdeepwiki.com\u002FPaddlePaddle\u002FPaddleOCR)\n[![AI Studio](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FPaddleOCR-_Offiical_Website-1927BA?logo=data:image\u002Fpng;base64,iVBORw0KGgoAAAANSUhEUgAAAgAAAAIACAMAAADDpiTIAAAABlBMVEU2P+X\u002F\u002F\u002F+1KuUwAAAHKklEQVR42u3dS5bjOAwEwALvf2fMavZum6IAImI7b2yYSqU+1Zb\u002F\u002FgAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAADKCR\u002F+fzly7rD92yVg69xh8zeLwOa5w+ZvFYHtc4ft3ykB++cOm79PAp6YO2z\u002FNgl4ZO5l+9+yT4QAvLqS748VF33Ylzdvzpl72f6z53YIGJ6SZdPeNHcIwOycaADdLgCSIgAIgCOAACAAykIAEAAEAAFAABCAT+WQuQVgeBqXhXQIQAAYegowLQBpbg3gZGFyAC6vgBQAMREA2\u002FYfDPxyaDQNyTNz+3Zwn5J4ZG7PB2h0kHhi7plPCImmJwkPzO0RMa3OET0i5uGlzHFze0xcu0vE2Dq3J4U2vEPgSaHbFzPNDQAAAAAAAMBNovdw+cP\u002Fny+uaf7w\u002F+eYADy8kE+F4Offdjn6zZXhAXgiA78G4MNNsmnu1Xr7b3mbOL8T5Ja5bw\u002FA35EC2LiWpzt1y9jRugBy30fLg3NvHPvnuZcC2NsCUXA\u002FaRmA89V07Fwgt37uH8deCmBr6N44pP4UgaUATpdA7v\u002FcMbIB8okliY65\u002FSW5HhJ1ehPmM+8edwXgpbu4R88FayR32Y\u002FP7oZZbOx13\u002FZr\u002F\u002FZHx27bAPnkFoyewYlbAhD3TvBobr95gaUAtr1EdNx1lgI4OcTTuR3z6+FZMEDRcu9ZCuDgGCdyGxMa4EgBRMvcjrkM7NgBZw5c0TwAUWUhZwRXA2xaya65Xa3jO2qYZ8bu2AD5w38tG5V8aZpoGN6Tz0bOfa9bceyWAciTO0jWyO1Tc5cLwJmF\u002FJfPnXVyu3\u002FslgHIg1n79O2O5fZv+1cHV7sC2HYqmUdHysNzX3sVkMcjUK5Gc+dMs28E5bGtm0V3gloBOP9vgZv+4sYn3RUaYFMCol5uN77g6lUApc8pWs69Zn7snS9Z9Q8G0S0AUTVUUTG3A54R1KSvo\u002FdiLAv5fKzynZeN6xogC75u93+AtBTA47OlAFSv6qY\u002Fvp3DAjD8iv2ZdFYJwKynMhTK1rInPfzaxW81LnvSgFP9KxrATaCLA3DxHpbFX31ZyNm5XRZyXG5bNkAWfP0rcrsUwOgC6NIAzgBcBiqAWwPgLrAGuGBP6jr2sifdfiJ6QQM4Bbw4AK4B3129ZSFn53ZZyA\u002FGyFty27IBFMDFAXAG8PbyLQv5xULGPRl0K3h2AbwcgCZPhs+LD1zLnjS6AN4NwMU\u002FDVFh7LyhASreTbvqrxdr\u002FJ4XT4Swz4FrTS+AGJ7bNbwAYkxuWzZAVljHrJfbjb9wviYXwFO\u002FFJ8Vli4vaICsEMFyBbA3tmtsAUS0zG1c\u002Fbj4YwsZH2\u002F+Whd0+1Nb+S7IE2sfPw4RL0XmsR8Nqvz7qFngmPHF34EqjP15AAofAkosZKPC\u002FK6FVoeP02Ehi540NG6AK\u002F4pYP3cLgVwXwHkDQ1QcSGb\u002FuF4WwCmfX8u\u002F+4vgLINcMUlQIfcLgXwXAF0+BGkpQDuuJx7\u002Fhwgpu\u002F\u002FcWVuO3wxJOz\u002Fz8297vgYBwaIO3O7Kn+c194578ltywbIgu8fl+Z2lS+APvnLjnOv8hsgSqxjgwL4Ln9LAezaj98tgPzy7ZcC+GQzxrWxXQpgx370dm6\u002FH7v6jaBoso5dY1swAFlwHWvfBf5pxVa93fCtdx64+1dsgCy4joWvAfPX9VoKYMs6Zse9\u002F8Mlvv7LILlhAfKFFdsSutJXAdFkL3qlADJPrXFcXAC5KYaH586jO9mtAch9S3T0GQJ726ZWAE49kjP3rlDJuetdaL\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\u002FgN+IH8U6YveYgAAAABJRU5ErkJggg==&labelColor=white)](https:\u002F\u002Fwww.paddleocr.com)\n\n\n\n**PaddleOCR 是一款业界领先的、可直接投入生产的 OCR（光学字符识别）和文档 AI 引擎，提供从文本提取到智能文档理解的端到端解决方案**\n\n\u003C\u002Fdiv>\n\n# PaddleOCR\n[![框架](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FPaddlePaddle-3.0-orange)](https:\u002F\u002Fwww.paddlepaddle.org.cn\u002Fen)\n[![准确率](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FRecognition%20Accuracy-🏆-green)](#)\n[![多语言](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FSupport_Languages-100+-brightgreen)](#)\n[![手写体](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FHandwriting-✓-success)](#)\n[![硬件](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FHeterogeneous%20Hardware-Kunlunxin%20%7C%20Ascend_NPU-red)](#)\n\n> [!TIP]\n> PaddleOCR 现已提供 MCP 服务器（Model Context Protocol），支持集成至 Claude Desktop 等智能体（Agent）应用。详情参见 [PaddleOCR MCP 服务器](https:\u002F\u002Fpaddlepaddle.github.io\u002FPaddleOCR\u002Flatest\u002Fen\u002Fversion3.x\u002Fdeployment\u002Fmcp_server.html)。\n>\n> PaddleOCR 3.0 技术报告现已发布。详见：[PaddleOCR 3.0 技术报告](https:\u002F\u002Farxiv.org\u002Fabs\u002F2507.05595)。\n>\n> PaddleOCR-VL 技术报告现已发布。详见 [PaddleOCR-VL 技术报告](https:\u002F\u002Farxiv.org\u002Fabs\u002F2510.14528)。\n>\n> PaddleOCR 官方网站 Beta 版现已上线，提供更便捷的在线体验及大规模 PDF 文件解析，并提供免费 API 和 MCP 服务。更多详情请访问 [PaddleOCR 官方网站](https:\u002F\u002Fwww.paddleocr.com)。\n\n\n**PaddleOCR** 将文档和图像转换为**结构化、AI 友好的数据**（如 JSON 和 Markdown），具备**业界领先的准确率**——为全球从独立开发者、初创公司到大型企业的各类用户构建 AI 应用赋能。拥有超过**60,000 个星标**，并深度集成于**MinerU, RAGFlow, pathway 和 cherry-studio**等领先项目中，PaddleOCR 已成为**AI 时代**开发者构建智能文档应用的**首选方案**。\n\n### PaddleOCR 3.0 核心功能\n\n[![官方网站](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FPaddleOCR--VL--1.5-_Official_Website-1927BA?logo=data:image\u002Fpng;base64,iVBORw0KGgoAAAANSUhEUgAAAgAAAAIACAMAAADDpiTIAAAABlBMVEU2P+X\u002F\u002F\u002F+1KuUwAAAHKklEQVR42u3dS5bjOAwEwALvf2fMavZum6IAImI7b2yYSqU+1Zb\u002F\u002FgAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAADKCR\u002F+fzly7rD92yVg69xh8zeLwOa5w+ZvFYHtc4ft3ykB++cOm79PAp6YO2z\u002FNgl4ZO5l+9+yT4QAvLqS748VF33Ylzdvzpl72f6z53YIGJ6SZdPeNHcIwOycaADdLgCSIgAIgCOAACAAykIAEAAEAAFAABCAT+WQuQVgeBqXhXQIQAAYegowLQBpbg3gZGFyAC6vgBQAMREA2\u002FYfDPxyaDQNyTNz+3Zwn5J4ZG7PB2h0kHhi7plPCImmJwkPzO0RMa3OET0i5uGlzHFze0xcu0vE2Dq3J4U2vEPgSaHbFzPNDQAAAAAAAMBNovdw+cP\u002Fny+uaf7w\u002F+eYADy8kE+F4Offdjn6zZXhAXgiA78G4MNNsmnu1Xr7b3mbOL8T5Ja5bw\u002FA35EC2LiWpzt1y9jRugBy30fLg3NvHPvnuZcC2NsCUXA\u002FaRmA89V07Fwgt37uH8deCmBr6N44pP4UgaUATpdA7v\u002FcMbIB8okliY65\u002FSW5HhJ1ehPmM+8edwXgpbu4R88FayR32Y\u002FP7oZZbOx13\u002FZr\u002F\u002FZHx27bAPnkFoyewYlbAhD3TvBobr95gaUAtr1EdNx1lgI4OcTTuR3z6+FZMEDRcu9ZCuDgGCdyGxMa4EgBRMvcjrkM7NgBZw5c0TwAUWUhZwRXA2xaya65Xa3jO2qYZ8bu2AD5w38tG5V8aZpoGN6Tz0bOfa9bceyWAciTO0jWyO1Tc5cLwJmF\u002FJfPnXVyu3\u002FslgHIg1n79O2O5fZv+1cHV7sC2HYqmUdHysNzX3sVkMcjUK5Gc+dMs28E5bGtm0V3gloBOP9vgZv+4sYn3RUaYFMCol5uN77g6lUApc8pWs69Zn7snS9Z9Q8G0S0AUTVUUTG3A54R1KSvo\u002FdiLAv5fKzynZeN6xogC75u93+AtBTA47OlAFSv6qY\u002Fvp3DAjD8iv2ZdFYJwKynMhTK1rInPfzaxW81LnvSgFP9KxrATaCLA3DxHpbFX31ZyNm5XRZyXG5bNkAWfP0rcrsUwOgC6NIAzgBcBiqAWwPgLrAGuGBP6jr2sifdfiJ6QQM4Bbw4AK4B3129ZSFn53ZZyA\u002FGyFty27IBFMDFAXAG8PbyLQv5xULGPRl0K3h2AbwcgCZPhs+LD1zLnjS6AN4NwMU\u002FDVFh7LyhASreTbvqrxdr\u002FJ4XT4Swz4FrTS+AGJ7bNbwAYkxuWzZAVljHrJfbjb9wviYXwFO\u002FFJ8Vli4vaICsEMFyBbA3tmtsAUS0zG1c\u002Fbj4YwsZH2\u002F+Whd0+1Nb+S7IE2sfPw4RL0XmsR8Nqvz7qFngmPHF34EqjP15AAofAkosZKPC\u002FK6FVoeP02Ehi540NG6AK\u002F4pYP3cLgVwXwHkDQ1QcSGb\u002FuF4WwCmfX8u\u002F+4vgLINcMUlQIfcLgXwXAF0+BGkpQDuuJx7\u002Fhwgpu\u002F\u002FcWVuO3wxJOz\u002Fz8297vgYBwaIO3O7Kn+c194578ltywbIgu8fl+Z2lS+APvnLjnOv8hsgSqxjgwL4Ln9LAezaj98tgPzy7ZcC+GQzxrWxXQpgx370dm6\u002FH7v6jaBoso5dY1swAFlwHWvfBf5pxVa93fCtdx64+1dsgCy4joWvAfPX9VoKYMs6Zse9\u002F8Mlvv7LILlhAfKFFdsSutJXAdFkL3qlADJPrXFcXAC5KYaH586jO9mtAch9S3T0GQJ726ZWAE49kjP3rlDJuetdaL\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\u002FgN+IH8U6YveYgAAAABJRU5ErkJggg==&labelColor=white)](https:\u002F\u002Faistudio.baidu.com\u002Fpaddleocr)\n[![HuggingFace](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FPaddleOCR--VL--1.5-_Demo_on_HuggingFace-yellow?logo=data:image\u002Fpng;base64,iVBORw0KGgoAAAANSUhEUgAAAF8AAABYCAMAAACkl9t\u002FAAAAk1BMVEVHcEz\u002FnQv\u002FnQv\u002FnQr\u002FnQv\u002FnQr\u002FnQv\u002FnQv\u002FnQr\u002FwRf\u002FtxT\u002Fpg7\u002FyRr\u002FrBD\u002FzRz\u002Fngv\u002FoAz\u002Fzhz\u002Fnwv\u002FtxT\u002Fngv\u002F0B3+zBz\u002FnQv\u002F0h7\u002Fwxn\u002FvRb\u002FthXkuiT\u002FrxH\u002FpxD\u002Fogzcqyf\u002FnQvTlSz\u002FczCxky7\u002FSjifdjT\u002FMj3+Mj3wMj15aTnDNz+DSD9RTUBsP0FRO0Q6O0WyIxEIAAAAGHRSTlMADB8zSWF3krDDw8TJ1NbX5efv8ff9\u002FfxKDJ9uAAAGKklEQVR42u2Z63qjOAyGC4RwCOfB2JAGqrSb2WnTw\u002F1f3UaWcSGYNKTdf\u002FP+mOkTrE+yJBulvfvLT2A5ruenaVHyIks33npl\u002F6C4s\u002FZLAM45SOi\u002F1FtZPyFur1OYofBX3w7d54Bxm+E8db+nDr12ttmESZ4zludJEG5S7TO72YPlKZFyE+YCYUJTBZsMiNS5Sd7NlDmKM2Eg2JQg8awbglfqgbhArjxkS7dgp2RH6hc9AMLdZYUtZN5DJr4molC8BfKrEkPKEnEVjLbgW1fLy77ZVOJagoIcLIl+IxaQZGjiX597HopF5CkaXVMDO9Pyix3AFV3kw4lQLCbHuMovz8FallbcQIJ5Ta0vks9RnolbCK84BtjKRS5uA43hYoZcOBGIG2Epbv6CvFVQ8m8loh66WNySsnN7htL58LNp+NXT8\u002FPhXiBXPMjLSxtwp8W9f\u002F1AngRierBkA+kk\u002FIpUSOeKByzn8y3kAAAfh\u002F\u002F0oXgV4roHm\u002Fkz4E2z\u002F\u002FzRc3\u002FlgwBzbM2mJxQEa5pqgX7d1L0htrhx7LKxOZlKbwcAWyEOWqYSI8YPtgDQVjpB5nvaHaSnBaQSD6hweDi8PosxD6\u002FPT09YY3xQA7LTCTKfYX+QHpA0GCcqmEHvr\u002FcyfKQTEuwgbs2kPxJEB0iNjfJcCTPyocx+A0griHSmADiC91oNGVwJ69RudYe65vJmoqfpul0lrqXadW0jFKH5BKwAeCq+Den7s+3zfRJzA61\u002FUj\u002F9H\u002FVzLKTx9jFPPdXeeP+L7WEvDLAKAIoF8bPTKT0+TM7W8ePj3Rz\u002FYn3kOAp2f1Kf0Weury7pn\u002FcPydvhQYV+eFOfmOu7VB\u002FViPe34\u002FEN3RFHY\u002FyRuT8ddCtMPH\u002FMcBAT5s+vRde\u002Fgf2c\u002FsPsjLK+m5IBQF5tO+h2tTlBGnP6693JdsvofjOPnnEHkh2TnV\u002FX1fBl9S5zrwuwF8NFrAVJVwCAPTe8gaJlomqlp0pv4Pjn98tJ\u002Ft\u002FfL++6unpR1YGC2n\u002FKCoa0tTLoKiEeUPDl94nj+5\u002FTv3\u002FeT5vBQ60X1S0oZr+IWRR8Ldhu7AlLjPISlJcO9vrFotky9SpzDequlwEir5beYAc0R7D9KS1DXva0jhYRDXoExPdc6yw5GShkZXe9QdO\u002FuOvHofxjrV\u002FTNS6iMJS+4TcSTgk9n5agJdBQbB\u002F\u002FIfF\u002FHpvPt3Tbi7b6I6K0R72p6ajryEJrENW2bbeVUGjfgoals4L443c7BEE4mJO2SpbRngxQrAKRudRzGQ8jVOL2qDVjjI8K1gc3TIJ5KiFZ1q+gdsARPB4NQS4AjwVSt72DSoXNyOWUrU5mQ9nRYyjp89Xo7oRI6Bga9QNT1mQ\u002FptaJq5T\u002F7WcgAZywR\u002FXlPGAUDdet3LE+qS0TI+g+aJU8MIqjo0Kx8Ly+maxLjJmjQ18rA0YCkxLQbUZP1WqdmyQGJLUm7VnQFqodmXSqmRrdVpqdzk5LvmvgtEcW8PMGdaS23EOWyDVbACZzUJPaqMbjDxpA3Qrgl0AikimGDbqmyT8P8NOYiqrldF8rX+YN7TopX4UoHuSCYY7cgX4gHwclQKl1zhx0THf+tCAUValzjI7Wg9EhptrkIcfIJjA94evOn8B2eHaVzvBrnl2ig0So6hvPaz0IGcOvTHvUIlE2+prqAxLSQxZlU2stql1NqCCLdIiIN\u002Fi1DBEHUoElM9dBravbiAnKqgpi4IBkw+utSPIoBijDXJipSVV7MpOEJUAc5Qmm3BnUN+w3hteEieYKfRZSIUcXKMVf0u5wD4EwsUNVvZOtUT7A2GkffHjByWpHqvRBYrTV72a6j8zZ6W0DTE86Hn04bmyWX3Ri9WH7ZU6Q7h+ZHo0nHUAcsQvVhXRDZHChwiyi\u002FhnPuOsSEF6Exk3o6Y9DT1eZ+6cASXk2Y9k+6EOQMDGm6WBK10wOQJCBwren86cPPWUcRAnTVjGcU1LBgs9FURiX\u002Fe6479yZcLwCBmTxiawEwrOcleuu12t3tbLv\u002FN4RLYIBhYexm7Fcn4OJcn0+zc+s8\u002FVfPeddZHAGN6TT8eGczHdR\u002FGts1\u002FMzDkThr23zqrVfAMFT33Nx1RJsx1k5zuWILLnG\u002FvsH+Fv5D4NTVcp1Gzo8AAAAAElFTkSuQmCC&labelColor=white)](https:\u002F\u002Fhuggingface.co\u002Fspaces\u002FPaddlePaddle\u002FPaddleOCR-VL-1.5_Online_Demo)\n[![ModelScope](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FPaddleOCR--VL--1.5-_Demo_on_ModelScope-purple?logo=data:image\u002Fsvg+xml;base64,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&labelColor=white)](https:\u002F\u002Fwww.modelscope.cn\u002Fstudios\u002FPaddlePaddle\u002FPaddleOCR-VL-1.5_Online_Demo)\n\n[![AI Studio（人工智能工作室）](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FPP--OCRv5-Demo_on_AI_Studio-1927BA?logo=data:image\u002Fpng;base64,iVBORw0KGgoAAAANSUhEUgAAAgAAAAIACAMAAADDpiTIAAAABlBMVEU2P+X\u002F\u002F\u002F+1KuUwAAAHKklEQVR42u3dS5bjOAwEwALvf2fMavZum6IAImI7b2yYSqU+1Zb\u002F\u002FgAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAADKCR\u002F+fzly7rD92yVg69xh8zeLwOa5w+ZvFYHtc4ft3ykB++cOm79PAp6YO2z\u002FNgl4ZO5l+9+yT4QAvLqS748VF33Ylzdvzpl72f6z53YIGJ6SZdPeNHcIwOycaADdLgCSIgAIgCOAACAAykIAEAAEAAFAABCAT+WQuQVgeBqXhXQIQAAYegowLQBpbg3gZGFyAC6vgBQAMREA2\u002FYfDPxyaDQNyTNz+3Zwn5J4ZG7PB2h0kHhi7plPCImmJwkPzO0RMa3OET0i5uGlzHFze0xcu0vE2Dq3J4U2vEPgSaHbFzPNDQAAAAAAAMBNovdw+cP\u002Fny+uaf7w\u002F+eYADy8kE+F4Offdjn6zZXhAXgiA78G4MNNsmnu1Xr7b3mbOL8T5Ja5bw\u002FA35EC2LiWpzt1y9jRugBy30fLg3NvHPvnuZcC2NsCUXA\u002FaRmA89V07Fwgt37uH8deCmBr6N44pP4UgaUATpdA7v\u002FcMbIB8okliY65\u002FSW5HhJ1ehPmM+8edwXgpbu4R88FayR32Y\u002FP7oZZbOx13\u002FZr\u002F\u002FZHx27bAPnkFoyewYlbAhD3TvBobr95gaUAtr1EdNx1lgI4OcTTuR3z6+FZMEDRcu9ZCuDgGCdyGxMa4EgBRMvcjrkM7NgBZw5c0TwAUWUhZwRXA2xaya65Xa3jO2qYZ8bu2AD5w38tG5V8aZpoGN6Tz0bOfa9bceyWAciTO0jWyO1Tc5cLwJmF\u002FJfPnXVyu3\u002FslgHIg1n79O2O5fZv+1cHV7sC2HYqmUdHysNzX3sVkMcjUK5Gc+dMs28E5bGtm0V3gloBOP9vgZv+4sYn3RUaYFMCol5uN77g6lUApc8pWs69Zn7snS9Z9Q8G0S0AUTVUUTG3A54R1KSvo\u002FdiLAv5fKzynZeN6xogC75u93+AtBTA47OlAFSv6qY\u002Fvp3DAjD8iv2ZdFYJwKynMhTK1rInPfzaxW81LnvSgFP9KxrATaCLA3DxHpbFX31ZyNm5XRZyXG5bNkAWfP0rcrsUwOgC6NIAzgBcBiqAWwPgLrAGuGBP6jr2sifdfiJ6QQM4Bbw4AK4B3129ZSFn53ZZyA\u002FGyFty27IBFMDFAXAG8PbyLQv5xULGPRl0K3h2AbwcgCZPhs+LD1zLnjS6AN4NwMU\u002FDVFh7LyhASreTbvqrxdr\u002FJ4XT4Swz4FrTS+AGJ7bNbwAYkxuWzZAVljHrJfbjb9wviYXwFO\u002FFJ8Vli4vaICsEMFyBbA3tmtsAUS0zG1c\u002Fbj4YwsZH2\u002F+Whd0+1Nb+S7IE2sfPw4RL0XmsR8Nqvz7qFngmPHF34EqjP15AAofAkosZKPC\u002FK6FVoeP02Ehi540NG6AK\u002F4pYP3cLgVwXwHkDQ1QcSGb\u002FuF4WwCmfX8u\u002F+4vgLINcMUlQIfcLgXwXAF0+BGkpQDuuJx7\u002Fhwgpu\u002F\u002FcWVuO3wxJOz\u002Fz8297vgYBwaIO3O7Kn+c194578ltywbIgu8fl+Z2lS+APvnLjnOv8hsgSqxjgwL4Ln9LAezaj98tgPzy7ZcC+GQzxrWxXQpgx370dm6\u002FH7v6jaBoso5dY1swAFlwHWvfBf5pxVa93fCtdx64+1dsgCy4joWvAfPX9VoKYMs6Zse9\u002F8Mlvv7LILlhAfKFFdsSutJXAdFkL3qlADJPrXFcXAC5KYaH586jO9mtAch9S3T0GQJ726ZWAE49kjP3rlDJuetdaL\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\u002FgN+IH8U6YveYgAAAABJRU5ErkJggg==&labelColor=white)](https:\u002F\u002Faistudio.baidu.com\u002Fcommunity\u002Fapp\u002F91660\u002FwebUI)\n[![AI Studio](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FPP--StructureV3-Demo_on_AI_Studio-1927BA?logo=data:image\u002Fpng;base64,iVBORw0KGgoAAAANSUhEUgAAAgAAAAIACAMAAADDpiTIAAAABlBMVEU2P+X\u002F\u002F\u002F+1KuUwAAAHKklEQVR42u3dS5bjOAwEwALvf2fMavZum6IAImI7b2yYSqU+1Zb\u002F\u002FgAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAADKCR\u002F+fzly7rD92yVg69xh8zeLwOa5w+ZvFYHtc4ft3ykB++cOm79PAp6YO2z\u002FNgl4ZO5l+9+yT4QAvLqS748VF33Ylzdvzpl72f6z53YIGJ6SZdPeNHcIwOycaADdLgCSIgAIgCOAACAAykIAEAAEAAFAABCAT+WQuQVgeBqXhXQIQAAYegowLQBpbg3gZGFyAC6vgBQAMREA2\u002FYfDPxyaDQNyTNz+3Zwn5J4ZG7PB2h0kHhi7plPCImmJwkPzO0RMa3OET0i5uGlzHFze0xcu0vE2Dq3J4U2vEPgSaHbFzPNDQAAAAAAAMBNovdw+cP\u002Fny+uaf7w\u002F+eYADy8kE+F4Offdjn6zZXhAXgiA78G4MNNsmnu1Xr7b3mbOL8T5Ja5bw\u002FA35EC2LiWpzt1y9jRugBy30fLg3NvHPvnuZcC2NsCUXA\u002FaRmA89V07Fwgt37uH8deCmBr6N44pP4UgaUATpdA7v\u002FcMbIB8okliY65\u002FSW5HhJ1ehPmM+8edwXgpbu4R88FayR32Y\u002FP7oZZbOx13\u002FZr\u002F\u002FZHx27bAPnkFoyewYlbAhD3TvBobr95gaUAtr1EdNx1lgI4OcTTuR3z6+FZMEDRcu9ZCuDgGCdyGxMa4EgBRMvcjrkM7NgBZw5c0TwAUWUhZwRXA2xaya65Xa3jO2qYZ8bu2AD5w38tG5V8aZpoGN6Tz0bOfa9bceyWAciTO0jWyO1Tc5cLwJmF\u002FJfPnXVyu3\u002FslgHIg1n79O2O5fZv+1cHV7sC2HYqmUdHysNzX3sVkMcjUK5Gc+dMs28E5bGtm0V3gloBOP9vgZv+4sYn3RUaYFMCol5uN77g6lUApc8pWs69Zn7snS9Z9Q8G0S0AUTVUUTG3A54R1KSvo\u002FdiLAv5fKzynZeN6xogC75u93+AtBTA47OlAFSv6qY\u002Fvp3DAjD8iv2ZdFYJwKynMhTK1rInPfzaxW81LnvSgFP9KxrATaCLA3DxHpbFX31ZyNm5XRZyXG5bNkAWfP0rcrsUwOgC6NIAzgBcBiqAWwPgLrAGuGBP6jr2sifdfiJ6QQM4Bbw4AK4B3129ZSFn53ZZyA\u002FGyFty27IBFMDFAXAG8PbyLQv5xULGPRl0K3h2AbwcgCZPhs+LD1zLnjS6AN4NwMU\u002FDVFh7LyhASreTbvqrxdr\u002FJ4XT4Swz4FrTS+AGJ7bNbwAYkxuWzZAVljHrJfbjb9wviYXwFO\u002FFJ8Vli4vaICsEMFyBbA3tmtsAUS0zG1c\u002Fbj4YwsZH2\u002F+Whd0+1Nb+S7IE2sfPw4RL0XmsR8Nqvz7qFngmPHF34EqjP15AAofAkosZKPC\u002FK6FVoeP02Ehi540NG6AK\u002F4pYP3cLgVwXwHkDQ1QcSGb\u002FuF4WwCmfX8u\u002F+4vgLINcMUlQIfcLgXwXAF0+BGkpQDuuJx7\u002Fhwgpu\u002F\u002FcWVuO3wxJOz\u002Fz8297vgYBwaIO3O7Kn+c194578ltywbIgu8fl+Z2lS+APvnLjnOv8hsgSqxjgwL4Ln9LAezaj98tgPzy7ZcC+GQzxrWxXQpgx370dm6\u002FH7v6jaBoso5dY1swAFlwHWvfBf5pxVa93fCtdx64+1dsgCy4joWvAfPX9VoKYMs6Zse9\u002F8Mlvv7LILlhAfKFFdsSutJXAdFkL3qlADJPrXFcXAC5KYaH586jO9mtAch9S3T0GQJ726ZWAE49kjP3rlDJuetdaL\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\u002FgN+IH8U6YveYgAAAABJRU5ErkJggg==&labelColor=white)](https:\u002F\u002Faistudio.baidu.com\u002Fcommunity\u002Fapp\u002F518494\u002FwebUI)\n[![AI Studio](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FPP--ChatOCRv4-Demo_on_AI_Studio-1927BA?logo=data:image\u002Fpng;base64,iVBORw0KGgoAAAANSUhEUgAAAgAAAAIACAMAAADDpiTIAAAABlBMVEU2P+X\u002F\u002F\u002F+1KuUwAAAHKklEQVR42u3dS5bjOAwEwALvf2fMavZum6IAImI7b2yYSqU+1Zb\u002F\u002FgAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAADKCR\u002F+fzly7rD92yVg69xh8zeLwOa5w+ZvFYHtc4ft3ykB++cOm79PAp6YO2z\u002FNgl4ZO5l+9+yT4QAvLqS748VF33Ylzdvzpl72f6z53YIGJ6SZdPeNHcIwOycaADdLgCSIgAIgCOAACAAykIAEAAEAAFAABCAT+WQuQVgeBqXhXQIQAAYegowLQBpbg3gZGFyAC6vgBQAMREA2\u002FYfDPxyaDQNyTNz+3Zwn5J4ZG7PB2h0kHhi7plPCImmJwkPzO0RMa3OET0i5uGlzHFze0xcu0vE2Dq3J4U2vEPgSaHbFzPNDQAAAAAAAMBNovdw+cP\u002Fny+uaf7w\u002F+eYADy8kE+F4Offdjn6zZXhAXgiA78G4MNNsmnu1Xr7b3mbOL8T5Ja5bw\u002FA35EC2LiWpzt1y9jRugBy30fLg3NvHPvnuZcC2NsCUXA\u002FaRmA89V07Fwgt37uH8deCmBr6N44pP4UgaUATpdA7v\u002FcMbIB8okliY65\u002FSW5HhJ1ehPmM+8edwXgpbu4R88FayR32Y\u002FP7oZZbOx13\u002FZr\u002F\u002FZHx27bAPnkFoyewYlbAhD3TvBobr95gaUAtr1EdNx1lgI4OcTTuR3z6+FZMEDRcu9ZCuDgGCdyGxMa4EgBRMvcjrkM7NgBZw5c0TwAUWUhZwRXA2xaya65Xa3jO2qYZ8bu2AD5w38tG5V8aZpoGN6Tz0bOfa9bceyWAciTO0jWyO1Tc5cLwJmF\u002FJfPnXVyu3\u002FslgHIg1n79O2O5fZv+1cHV7sC2HYqmUdHysNzX3sVkMcjUK5Gc+dMs28E5bGtm0V3gloBOP9vgZv+4sYn3RUaYFMCol5uN77g6lUApc8pWs69Zn7snS9Z9Q8G0S0AUTVUUTG3A54R1KSvo\u002FdiLAv5fKzynZeN6xogC75u93+AtBTA47OlAFSv6qY\u002Fvp3DAjD8iv2ZdFYJwKynMhTK1rInPfzaxW81LnvSgFP9KxrATaCLA3DxHpbFX31ZyNm5XRZyXG5bNkAWfP0rcrsUwOgC6NIAzgBcBiqAWwPgLrAGuGBP6jr2sifdfiJ6QQM4Bbw4AK4B3129ZSFn53ZZyA\u002FGyFty27IBFMDFAXAG8PbyLQv5xULGPRl0K3h2AbwcgCZPhs+LD1zLnjS6AN4NwMU\u002FDVFh7LyhASreTbvqrxdr\u002FJ4XT4Swz4FrTS+AGJ7bNbwAYkxuWzZAVljHrJfbjb9wviYXwFO\u002FFJ8Vli4vaICsEMFyBbA3tmtsAUS0zG1c\u002Fbj4YwsZH2\u002F+Whd0+1Nb+S7IE2sfPw4RL0XmsR8Nqvz7qFngmPHF34EqjP15AAofAkosZKPC\u002FK6FVoeP02Ehi540NG6AK\u002F4pYP3cLgVwXwHkDQ1QcSGb\u002FuF4WwCmfX8u\u002F+4vgLINcMUlQIfcLgXwXAF0+BGkpQDuuJx7\u002Fhwgpu\u002F\u002FcWVuO3wxJOz\u002Fz8297vgYBwaIO3O7Kn+c194578ltywbIgu8fl+Z2lS+APvnLjnOv8hsgSqxjgwL4Ln9LAezaj98tgPzy7ZcC+GQzxrWxXQpgx370dm6\u002FH7v6jaBoso5dY1swAFlwHWvfBf5pxVa93fCtdx64+1dsgCy4joWvAfPX9VoKYMs6Zse9\u002F8Mlvv7LILlhAfKFFdsSutJXAdFkL3qlADJPrXFcXAC5KYaH586jO9mtAch9S3T0GQJ726ZWAE49kjP3rlDJuetdaL\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\u002FgN+IH8U6YveYgAAAABJRU5ErkJggg==&labelColor=white)](https:\u002F\u002Faistudio.baidu.com\u002Fcommunity\u002Fapp\u002F518493\u002FwebUI)\n\n- **PaddleOCR-VL-1.5：面向真实世界文档解析与文本检测（Text Spotting）的 0.9B VLM（视觉语言模型）**\n  一款 SOTA（State-of-the-Art，最先进）且资源高效的模型，专为真实世界文档解析和文本检测任务设计。它在文档解析任务的六大主要场景中实现了全面领先：正常、倾斜、弯曲、扫描、光照变化和屏幕拍摄。它引入了领先的文本检测和印章识别新能力，加强了对复杂元素（如文本、表格、公式和图表）的解析，并将语言支持扩展到 111 种语言——同时保持极低的资源消耗。\n\n- **PaddleOCR-VL - 通过 0.9B VLM 实现多语言文档解析**  \n  **专为文档解析量身定制的 SOTA 和资源高效模型**，支持 109 种语言，擅长识别复杂元素（例如文本、表格、公式和图表），同时保持最低的资源消耗。\n\n- **PP-OCRv5 — 通用场景文本识别**  \n  **单模型支持五种文本类型**（简体中文、繁体中文、英文、日文和拼音），**准确率提升 13%**。解决多语言混合文档识别挑战。\n\n- **PP-StructureV3 — 复杂文档解析**  \n  智能地将复杂的 PDF 和文档图像转换为**保留原始结构的 Markdown 和 JSON 文件**。在公开基准测试中**优于**众多商业解决方案。**完美保持文档布局和层级结构**。\n\n- **PP-ChatOCRv4 — 智能信息提取**  \n  原生集成 ERNIE 4.5，从海量文档中**精确提取关键信息**，相比上一代准确率提升 15%。使文档能够\"**理解**\"您的问题并提供准确答案。\n\n除了提供出色的模型库外，PaddleOCR 3.0 还提供了用户友好的工具，涵盖模型训练、推理和服务部署，使开发人员能够快速将 AI 应用投入生产。\n\u003Cdiv align=\"center\">\n  \u003Cp>\n      \u003Cimg width=\"100%\" src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FPaddlePaddle_PaddleOCR_readme_603d058de9a9.jpg\" alt=\"PaddleOCR Architecture\">\n  \u003C\u002Fp>\n\u003C\u002Fdiv>\n\n**特别提示**：PaddleOCR 3.x 引入了若干重要的接口变更。**基于 PaddleOCR 2.x 编写的旧代码可能与 PaddleOCR 3.x 不兼容**。请确保您阅读的文档与您使用的 PaddleOCR 版本相匹配。[本文档](https:\u002F\u002Fpaddlepaddle.github.io\u002FPaddleOCR\u002Flatest\u002Fen\u002Fupdate\u002Fupgrade_notes.html) 解释了升级原因以及从 PaddleOCR 2.x 到 3.x 的主要变更。\n\n\n\n## 📣 近期更新\n\n### 🔥🔥 2026.1.29：发布 PaddleOCR 3.4.0，包含：\n\n- 发布 PaddleOCR-VL-1.5：\n    - **模型介绍：**\n      - PaddleOCR-VL-1.5 是 PaddleOCR-VL 的新版本，更加关注真实世界场景下的文档解析能力并扩展新功能。依托创新的 **PP-DocLayoutV3** 算法进行不规则形状定位，它在涉及倾斜、弯曲、扫描、光照变化和屏幕拍摄的自然文档场景中实现了精确的布局分析。紧凑的多模态模型 **PaddleOCR-VL-1.5-0.9B** 保持了原有的参数量，同时扩展了其在文本检测和印章识别方面的领先能力。语言支持已扩展至 **111 种语言**，复杂场景中的元素识别能力显著增强。该模型现已在 [HuggingFace](https:\u002F\u002Fhuggingface.co\u002FPaddlePaddle\u002FPaddleOCR-VL-1.5) 上提供。您也可以通过 [PaddleOCR 官方网站](https:\u002F\u002Fwww.paddleocr.com) 在线体验或调用 API。\n  - **核心功能：**\n    - **文档解析领域的 SOTA 性能：** PaddleOCR-VL-1.5 在 OmniDocBench v1.5 基准测试上达到了 **94.5%** 的高精度，超越了顶级全球通用大模型和专用文档解析模型。\n    - **五大真实世界场景的 SOTA 性能：** 引入创新的文档解析方法，它是首个支持不规则文档布局定位的模型。在所有五个真实世界场景中均优于开源和闭源模型：倾斜、弯曲、扫描、光照变化和屏幕拍摄。\n    - **基于 0.9B 紧凑模型的扩展能力：** 仅 0.9B 的参数量，PaddleOCR-VL-1.5 将其范围扩展至包括文本检测和印章识别，在这些相关任务中设定了新的 SOTA 结果。\n    - **增强的多元素识别：** 该模型针对特定场景和多语言内容的识别性能有所提升，包括特殊符号、古籍文本、多语言表格、下划线和复选框。语言覆盖范围已扩展至包括 **藏文和孟加拉文**。\n    - **长文档跨页解析：** 该模型支持跨页表格的自动合并以及跨页段落标题的识别，有效解决了长文档解析中的内容碎片化问题。\n\n\u003Cdetails>\n\u003Csummary>\u003Cstrong>2025.10.16：发布 PaddleOCR 3.3.0\u003C\u002Fstrong>\u003C\u002Fsummary>\n\n- 发布 PaddleOCR-VL：\n    - **模型介绍**：\n        - **PaddleOCR-VL** 是一款 SOTA 且资源高效的模型，专为文档解析量身定制。其核心组件是 PaddleOCR-VL-0.9B，这是一个紧凑而强大的视觉语言模型（VLM），集成了 NaViT 风格的动态分辨率视觉编码器与 ERNIE-4.5-0.3B 语言模型，以实现准确的元素识别。**这款创新模型高效支持 109 种语言，擅长识别复杂元素（例如文本、表格、公式和图表），同时保持最低的资源消耗**。通过在广泛使用的公开基准测试和内部基准测试上的综合评估，PaddleOCR-VL 在页面级文档解析和元素级识别方面均实现了 SOTA 性能。它显著优于现有解决方案，在与顶级 VLM 的竞争中表现出强劲的竞争力，并提供了快速的推理速度。这些优势使其非常适合在实际场景中进行部署。该模型已在 [HuggingFace](https:\u002F\u002Fhuggingface.co\u002FPaddlePaddle\u002FPaddleOCR-VL) 上发布。欢迎大家下载和使用！更多信息可在 [PaddleOCR-VL](https:\u002F\u002Fwww.paddleocr.ai\u002Flatest\u002Fversion3.x\u002Falgorithm\u002FPaddleOCR-VL\u002FPaddleOCR-VL.html) 中找到。\n\n- **核心功能**：\n        - **紧凑而强大的 VLM（视觉语言模型）架构**：我们提出了一种专为资源高效推理设计的新型视觉语言模型，在元素识别方面取得了卓越的性能。通过将 NaViT 风格的动态高分辨率视觉编码器与轻量级 ERNIE-4.5-0.3B 语言模型相结合，我们显著增强了模型的识别能力和解码效率。这种集成在降低计算需求的同时保持了高精度，使其非常适合高效且实用的文档处理应用。\n        - **文档解析的 SOTA（业界领先）性能**：PaddleOCR-VL 在页面级文档解析和元素级识别方面均实现了业界领先的性能。它显著优于现有的基于流水线的方法，并在文档解析方面展现出与领先的视觉语言模型（VLMs）竞争的强大实力。此外，它在识别复杂文档元素（如文本、表格、公式和图表）方面表现出色，使其适用于各种具有挑战性的内容类型，包括手写文本和历史文档。这使其具有高度通用性，适用于广泛的文档类型和场景。\n        - **多语言支持**：PaddleOCR-VL 支持 109 种语言，涵盖主要全球语言，包括但不限于中文、英文、日文、拉丁字母语言、韩文，以及具有不同书写系统和结构的语言，如俄语（西里尔字母）、阿拉伯语、印地语（天城文）和泰语。这种广泛的语言覆盖范围显著增强了我们的系统在多语言和全球化文档处理场景中的适用性。\n\n- 发布 PP-OCRv5 多语言识别模型：\n    - 提高了拉丁字母识别的准确率和覆盖率；增加了对西里尔字母、阿拉伯语、天城文、泰卢固语、泰米尔语等其他语言系统的支持，涵盖 109 种语言的识别。该模型仅拥有 200 万参数，部分模型的准确率相比上一代提升了超过 40%。\n\n\u003C\u002Fdetails>\n\n\n\u003Cdetails>\n\u003Csummary>\u003Cstrong>2025.08.21: PaddleOCR 3.2.0 发布\u003C\u002Fstrong>\u003C\u002Fsummary>\n\n- **重大模型新增：**\n    - 推出了 PP-OCRv5 英语、泰语和希腊语识别模型的训练、推理和部署。**PP-OCRv5 英语模型在英语场景下相比主 PP-OCRv5 模型提升了 11%，泰语和希腊语识别模型的准确率分别达到了 82.68% 和 89.28%。**\n\n- **部署能力升级：**\n    - **完全支持 PaddlePaddle 框架版本 3.1.0 和 3.1.1。**\n    - **全面升级 PP-OCRv5 C++ 本地部署方案，现同时支持 Linux 和 Windows，功能与 Python 实现保持一致，精度相同。**\n    - **高性能推理现在支持 CUDA 12，推理可使用 Paddle Inference 或 ONNX Runtime 后端进行。**\n    - **高稳定性面向服务的部署方案现已完全开源，允许用户根据需要自定义 Docker 镜像和 SDK。**\n    - 高稳定性面向服务的部署方案还支持通过手动构建的 HTTP 请求进行调用，使得可以使用任何编程语言开发客户端代码。\n\n- **基准测试支持：**\n    - **所有流水线现在都支持细粒度基准测试，能够测量端到端推理时间以及逐层和逐模块的延迟数据，以协助性能分析。[这里](docs\u002Fversion3.x\u002Fpipeline_usage\u002Finstructions\u002Fbenchmark.en.md) 介绍了如何设置和使用基准测试功能。**\n    - **文档已更新，包含主流硬件上常用配置的关键指标，如推理延迟和内存占用，为用户提供部署参考。**\n\n- **修复问题：**\n    - 解决了模型训练期间日志保存失败的问题。\n    - 升级了公式模型的数据增强组件以兼容较新版本的 albumentations 依赖项，并修复了在多进程场景中使用 tokenizers 包时的死锁警告。\n    - 修复了 PP-StructureV3 配置文件中与其他流水线相比开关行为不一致的问题（例如 `use_chart_parsing`）。\n\n- **其他增强：**\n    - **分离了核心和可选依赖项。基础文本识别仅需最小核心依赖；可根据需要安装用于文档解析和信息提取的其他依赖项。**\n    - **启用了 Windows 上 NVIDIA RTX 50 系列显卡的支持；用户可以参考 [安装指南](docs\u002Fversion3.x\u002Finstallation.en.md) 获取相应的 PaddlePaddle 框架版本。**\n    - **PP-OCR 系列模型现在支持返回单字符坐标。**\n    - 添加了 AIStudio、ModelScope 等其他模型下载源，允许用户指定模型下载来源。\n    - 添加了对通过 PP-Chart2Table 模块进行图表转表格转换的支持。\n    - 优化了文档描述以提高可用性。\n\u003C\u002Fdetails>\n\n\u003Cdetails>\n\u003Csummary>\u003Cstrong>2025.08.15: PaddleOCR 3.1.1 发布\u003C\u002Fstrong>\u003C\u002Fsummary>\n\n- **修复问题：**\n  - 在 `PP-ChatOCRv4` 类中添加了缺失的方法 `save_vector`、`save_visual_info_list`、`load_vector` 和 `load_visual_info_list`。\n  - 在 `PPDocTranslation` 类的 `translate` 方法中添加了缺失的参数 `glossary` 和 `llm_request_interval`。\n\n- **文档改进：**\n  - 在 MCP 文档中添加了演示。\n  - 在文档中添加了用于性能指标测试所使用的 PaddlePaddle 和 PaddleOCR 版本的信息。\n  - 修复了流水线文档翻译中的错误和遗漏。\n\n- **其他：**\n  - 将 MCP 服务器依赖更改为使用纯 Python 库 `puremagic` 而不是 `python-magic`，以减少安装问题。\n  - 使用 PaddleOCR 版本 3.1.0 重新测试了 PP-OCRv5 性能指标并更新了文档。\n\n\u003C\u002Fdetails>\n\n\u003Cdetails>\n\u003Csummary>\u003Cstrong>2025.06.29: PaddleOCR 3.1.0 发布\u003C\u002Fstrong>\u003C\u002Fsummary>\n\n- **关键模型与流程（Pipelines）：**\n  - **新增了 PP-OCRv5 多语言文本识别模型**，支持 37 种语言（包括法语、西班牙语、葡萄牙语、俄语、韩语等）文本识别模型的训练和推理过程。**平均准确率提升了 30% 以上。** [详情](https:\u002F\u002Fpaddlepaddle.github.io\u002FPaddleOCR\u002Flatest\u002Fen\u002Fversion3.x\u002Falgorithm\u002FPP-OCRv5\u002FPP-OCRv5_multi_languages.html)\n  - 升级了 **PP-StructureV3** 中的 **PP-Chart2Table 模型**，进一步增强了图表转表格的能力。在内部自定义评估集上，指标 (RMS-F1) **增加了 9.36 个百分点 (71.24% -> 80.60%)**。\n  - 全新推出了基于 **PP-StructureV3** 和 **ERNIE 4.5** 的 **文档翻译流程（PP-DocTranslation）**，支持 Markdown 格式文档、各种复杂布局 PDF 文档以及文档图像的翻译，并将结果保存为 Markdown 格式文档。[详情](https:\u002F\u002Fpaddlepaddle.github.io\u002FPaddleOCR\u002Flatest\u002Fen\u002Fversion3.x\u002Fpipeline_usage\u002FPP-DocTranslation.html)\n\n\n- **新的 MCP 服务器：** [详情](https:\u002F\u002Fpaddlepaddle.github.io\u002FPaddleOCR\u002Flatest\u002Fen\u002Fversion3.x\u002Fdeployment\u002Fmcp_server.html)\n  - **同时支持 OCR（光学字符识别）和 PP-StructureV3 流程。**\n  - 支持三种工作模式：本地 Python 库、AIStudio 社区云服务以及自托管服务。\n  - 支持通过 stdio 调用本地服务，通过 Streamable HTTP 调用远程服务。\n\n- **文档优化：** 改进了部分用户指南的描述，以获得更流畅的阅读体验。\n\n\u003C\u002Fdetails>\n\n\u003Cdetails>\n    \u003Csummary>\u003Cstrong>2025.06.26：PaddleOCR 3.0.3 发布\u003C\u002Fstrong>\u003C\u002Fsummary>\n- **问题修复：** 解决了 `enable_mkldnn` 参数未生效的问题，恢复了使用 MKL-DNN 进行 CPU 推理的默认行为。\n\u003C\u002Fdetails>\n\n\u003Cdetails>\n    \u003Csummary>\u003Cstrong>2025.06.19：PaddleOCR 3.0.2 发布\u003C\u002Fstrong>\u003C\u002Fsummary>\n- **新功能：**\n\n  - 默认下载源已从 `BOS` 更改为 `HuggingFace`。用户也可以将环境变量 `PADDLE_PDX_MODEL_SOURCE` 更改为 `BOS`，将模型下载源设置回百度对象存储 (BOS)。\n  - 为 PP-OCRv5、PP-StructureV3 和 PP-ChatOCRv4 等流程添加了六种语言（C++、Java、Go、C#、Node.js 和 PHP）的服务调用示例。\n  - 改进了 PP-StructureV3 流程中的布局分区排序算法，增强了复杂垂直布局的排序逻辑，以提供更好的结果。\n  - 增强了模型选择逻辑：当指定了语言但未指定模型版本时，系统将自动选择支持该语言的最新模型版本。 \n  - 设置了 MKL-DNN 缓存大小的默认上限以防止无限增长，同时也允许用户配置缓存容量。\n  - 更新了高性能推理的默认配置以支持 Paddle MKL-DNN 加速，并优化了自动配置选择的逻辑以实现更智能的选择。\n  - 调整了获取默认设备的逻辑，以考虑已安装的 Paddle 框架对计算设备的实际支持情况，使程序行为更加直观。\n  - 添加了 PP-OCRv5 的 Android 示例。[详情](https:\u002F\u002Fpaddlepaddle.github.io\u002FPaddleOCR\u002Flatest\u002Fen\u002Fversion3.x\u002Fdeployment\u002Fon_device_deployment.html)。\n\n- **问题修复：**\n  - 修复了 PP-StructureV3 中某些 CLI 参数未生效的问题。\n  - 解决了在某些情况下 `export_paddlex_config_to_yaml` 无法正常工作的问题。\n  - 纠正了 `save_path` 的实际行为与其文档描述之间的差异。\n  - 修复了在基础服务部署中使用 MKL-DNN 时的潜在多线程错误。\n  - 纠正了 Latex-OCR 模型图像预处理中的通道顺序错误。\n  - 修复了文本识别模块中保存可视化图像时的通道顺序错误。\n  - 解决了 PP-StructureV3 流程中可视化表格结果的通道顺序错误。\n  - 修复了 PP-StructureV3 流程中 `overlap_ratio` 计算在极端特殊情况下的溢出问题。\n\n- **文档改进：**\n  - 更新了文档中 `enable_mkldnn` 参数的描述，以准确反映程序的实际行为。\n  - 修复了关于 `lang` 和 `ocr_version` 参数的文档错误。\n  - 添加了通过 CLI 导出流程配置文件的说明。\n  - 修复了 PP-OCRv5 性能数据表中缺失的列。\n  - 完善了不同配置下 PP-StructureV3 的基准测试指标。\n\n- **其他：**\n\n  - 放宽了对 numpy 和 pandas 等依赖项的版本限制，恢复了对 Python 3.12 的支持。\n\u003C\u002Fdetails>\n\n\u003Cdetails>\n    \u003Csummary>\u003Cstrong>历史日志\u003C\u002Fstrong>\u003C\u002Fsummary>\n\n2025.06.05：**PaddleOCR 3.0.1 发布**，包含：\n\n- **部分模型及模型配置的优化：**\n  - 更新了 PP-OCRv5 的默认模型配置，将检测和识别均从移动端模型更改为服务端模型。为了在大多数场景中提高默认性能，配置中的参数 `limit_side_len` 已从 736 更改为 64。\n  - 新增了一个准确率为 99.42% 的文本行方向分类模型 `PP-LCNet_x1_0_textline_ori`。OCR、PP-StructureV3 和 PP-ChatOCRv4 流程的默认文本行方向分类器已更新为此模型。\n  - 优化了文本行方向分类模型 `PP-LCNet_x0_25_textline_ori`，准确率提升了 3.3 个百分点，当前准确率为 98.85%。\n- **针对 3.0.0 版本部分问题的优化与修复，[详情](https:\u002F\u002Fpaddlepaddle.github.io\u002FPaddleOCR\u002Flatest\u002Fen\u002Fupdate\u002Fupdate.html)**\n\n🔥🔥2025.05.20：**PaddleOCR v3.0** 正式发布，包括：\n- **PP-OCRv5**：全场景高精度文本识别模型 - 图片\u002FPDF 即时提取文本。\n   1. 🌐 单模型支持**五种**文本类型 - 在单个模型内无缝处理**简体中文、繁体中文、简体中文拼音、英语**和**日语**。\n   2. ✍️ 改进了**手写体识别**：在处理复杂草书和非标准手写体方面显著更好。\n   3. 🎯 相比 PP-OCRv4 **准确率提升 13 个点**，在各种真实场景中实现了最先进的性能（SOTA）。\n\n- **PP-StructureV3**：通用文档解析 – 释放 SOTA（最先进）图片\u002FPDF 解析能力以应对真实场景！ \n   1. 🧮 **高精度多场景 PDF 解析**，在 OmniDocBench 基准测试中领先开源和闭源解决方案。\n   2. 🧠 专项能力包括**印章识别**、**图表转表格**、**带嵌套公式\u002F图像的表格识别**、**竖排文档解析**和**复杂表格结构分析**。\n\n- **PP-ChatOCRv4**: 智能文档理解——从图像\u002FPDF 中提取关键信息，而不仅仅是文本。\n   1. 🔥 在 PDF\u002FPNG\u002FJPG 文件的关键信息提取方面，相比上一代提升了 **15 个百分点的准确率**。\n   2. 💻 原生支持 **ERNIE 4.5**（文心），并兼容通过 PaddleNLP、Ollama、vLLM 等进行的大模型部署。\n   3. 🤝 集成 [PP-DocBee2](https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleMIX\u002Ftree\u002Fdevelop\u002Fpaddlemix\u002Fexamples\u002Fppdocbee2)，支持提取和理解复杂文档中的印刷文本、手写体、印章、表格、图表等常见元素。\n\n[更新日志](https:\u002F\u002Fpaddlepaddle.github.io\u002FPaddleOCR\u002Flatest\u002Fen\u002Fupdate\u002Fupdate.html)\n\n\u003C\u002Fdetails>\n\n\n\n## ⚡ 快速开始\n### 1. 运行在线演示 \n[![AI Studio](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FPP_OCRv5-AI_Studio-green)](https:\u002F\u002Faistudio.baidu.com\u002Fcommunity\u002Fapp\u002F91660\u002FwebUI)\n[![AI Studio](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FPP_StructureV3-AI_Studio-green)](https:\u002F\u002Faistudio.baidu.com\u002Fcommunity\u002Fapp\u002F518494\u002FwebUI)\n[![AI Studio](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FPP_ChatOCRv4-AI_Studio-green)](https:\u002F\u002Faistudio.baidu.com\u002Fcommunity\u002Fapp\u002F518493\u002FwebUI)\n\n### 2. 安装\n\n请参考 [安装指南](https:\u002F\u002Fwww.paddlepaddle.org.cn\u002Fen\u002Finstall\u002Fquick?docurl=\u002Fdocumentation\u002Fdocs\u002Fen\u002Fdevelop\u002Finstall\u002Fpip\u002Flinux-pip_en.html) 安装 PaddlePaddle（飞桨），之后安装 PaddleOCR 工具包。\n\n```bash\n# If you only want to use the basic text recognition feature (returns text position coordinates and content), including the PP-OCR series\npython -m pip install paddleocr\n# If you want to use all features such as document parsing, document understanding, document translation, key information extraction, etc.\n# python -m pip install \"paddleocr[all]\"\n```\n\n从 3.2.0 版本开始，除了上述演示的 `all` 依赖组外，PaddleOCR 还支持通过指定其他依赖组来安装部分可选功能。PaddleOCR 提供的所有依赖组如下：\n\n| 依赖组名称 | 对应功能 |\n| - | - |\n| `doc-parser` | 文档解析：可用于从文档中提取布局元素，如表格、公式、印章、图片等；包含 PP-StructureV3、PaddleOCR-VL 等模型 |\n| `ie` | 信息抽取：可用于从文档中提取关键信息，如姓名、日期、地址、金额等；包含 PP-ChatOCRv4 等模型 |\n| `trans` | 文档翻译：可用于将文档从一种语言翻译成另一种语言；包含 PP-DocTranslation 等模型 |\n| `all` | 完整功能 |\n\n### 3. 通过 CLI（命令行接口）运行推理\n```bash\n# Run PP-OCRv5 inference\npaddleocr ocr -i https:\u002F\u002Fpaddle-model-ecology.bj.bcebos.com\u002Fpaddlex\u002Fimgs\u002Fdemo_image\u002Fgeneral_ocr_002.png --use_doc_orientation_classify False --use_doc_unwarping False --use_textline_orientation False  \n\n# Run PP-StructureV3 inference\npaddleocr pp_structurev3 -i https:\u002F\u002Fpaddle-model-ecology.bj.bcebos.com\u002Fpaddlex\u002Fimgs\u002Fdemo_image\u002Fpp_structure_v3_demo.png --use_doc_orientation_classify False --use_doc_unwarping False\n\n# Get the Qianfan API Key at first, and then run PP-ChatOCRv4 inference\npaddleocr pp_chatocrv4_doc -i https:\u002F\u002Fpaddle-model-ecology.bj.bcebos.com\u002Fpaddlex\u002Fimgs\u002Fdemo_image\u002Fvehicle_certificate-1.png -k 驾驶室准乘人数 --qianfan_api_key your_api_key --use_doc_orientation_classify False --use_doc_unwarping False \n\n# Run PaddleOCR-VL inference\npaddleocr doc_parser -i https:\u002F\u002Fpaddle-model-ecology.bj.bcebos.com\u002Fpaddlex\u002Fimgs\u002Fdemo_image\u002Fpaddleocr_vl_demo.png\n\n# Get more information about \"paddleocr ocr\"\npaddleocr ocr --help\n```\n\n### 4. 通过 API（应用程序编程接口）运行推理\n**4.1 PP-OCRv5 示例**\n```python\n# Initialize PaddleOCR instance\nfrom paddleocr import PaddleOCR\nocr = PaddleOCR(\n    use_doc_orientation_classify=False,\n    use_doc_unwarping=False,\n    use_textline_orientation=False)\n\n# Run OCR inference on a sample image \nresult = ocr.predict(\n    input=\"https:\u002F\u002Fpaddle-model-ecology.bj.bcebos.com\u002Fpaddlex\u002Fimgs\u002Fdemo_image\u002Fgeneral_ocr_002.png\")\n\n# Visualize the results and save the JSON results\nfor res in result:\n    res.print()\n    res.save_to_img(\"output\")\n    res.save_to_json(\"output\")\n```\n\n\u003Cdetails>\n    \u003Csummary>\u003Cstrong>4.2 PP-StructureV3 示例\u003C\u002Fstrong>\u003C\u002Fsummary>\n\n```python\nfrom pathlib import Path\nfrom paddleocr import PPStructureV3\n\npipeline = PPStructureV3(\n    use_doc_orientation_classify=False,\n    use_doc_unwarping=False\n)\n\n# For Image\noutput = pipeline.predict(\n    input=\"https:\u002F\u002Fpaddle-model-ecology.bj.bcebos.com\u002Fpaddlex\u002Fimgs\u002Fdemo_image\u002Fpp_structure_v3_demo.png\",\n)\n\n# Visualize the results and save the JSON results\nfor res in output:\n    res.print() \n    res.save_to_json(save_path=\"output\") \n    res.save_to_markdown(save_path=\"output\")           \n```\n\n\u003C\u002Fdetails>\n\n\u003Cdetails>\n   \u003Csummary>\u003Cstrong>4.3 PP-ChatOCRv4 示例\u003C\u002Fstrong>\u003C\u002Fsummary>\n\n```python\nfrom paddleocr import PPChatOCRv4Doc\n\nchat_bot_config = {\n    \"module_name\": \"chat_bot\",\n    \"model_name\": \"ernie-3.5-8k\",\n    \"base_url\": \"https:\u002F\u002Fqianfan.baidubce.com\u002Fv2\",\n    \"api_type\": \"openai\",\n    \"api_key\": \"api_key\",  # your api_key\n}\n\nretriever_config = {\n    \"module_name\": \"retriever\",\n    \"model_name\": \"embedding-v1\",\n    \"base_url\": \"https:\u002F\u002Fqianfan.baidubce.com\u002Fv2\",\n    \"api_type\": \"qianfan\",\n    \"api_key\": \"api_key\",  # your api_key\n}\n\npipeline = PPChatOCRv4Doc(\n    use_doc_orientation_classify=False,\n    use_doc_unwarping=False\n)\n\nvisual_predict_res = pipeline.visual_predict(\n    input=\"https:\u002F\u002Fpaddle-model-ecology.bj.bcebos.com\u002Fpaddlex\u002Fimgs\u002Fdemo_image\u002Fvehicle_certificate-1.png\",\n    use_common_ocr=True,\n    use_seal_recognition=True,\n    use_table_recognition=True,\n)\n\nmllm_predict_info = None\nuse_mllm = False\n```\n\n# 如果使用多模态大模型（Multimodal Large Model），则需要启动本地多模态大语言模型（MLLM）服务。您可以参考文档：https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleX\u002Fblob\u002Frelease\u002F3.0\u002Fdocs\u002Fpipeline_usage\u002Ftutorials\u002Fvlm_pipelines\u002Fdoc_understanding.en.md 进行部署并更新 mllm_chat_bot_config 配置。\nif use_mllm:\n    mllm_chat_bot_config = {\n        \"module_name\": \"chat_bot\",\n        \"model_name\": \"PP-DocBee\",\n        \"base_url\": \"http:\u002F\u002F127.0.0.1:8080\u002F\",  # your local mllm service url\n        \"api_type\": \"openai\",\n        \"api_key\": \"api_key\",  # your api_key\n    }\n\n    mllm_predict_res = pipeline.mllm_pred(\n        input=\"https:\u002F\u002Fpaddle-model-ecology.bj.bcebos.com\u002Fpaddlex\u002Fimgs\u002Fdemo_image\u002Fvehicle_certificate-1.png\",\n        key_list=[\"驾驶室准乘人数\"],\n        mllm_chat_bot_config=mllm_chat_bot_config,\n    )\n    mllm_predict_info = mllm_predict_res[\"mllm_res\"]\n\nvisual_info_list = []\nfor res in visual_predict_res:\n    visual_info_list.append(res[\"visual_info\"])\n    layout_parsing_result = res[\"layout_parsing_result\"]\n\nvector_info = pipeline.build_vector(\n    visual_info_list, flag_save_bytes_vector=True, retriever_config=retriever_config\n)\nchat_result = pipeline.chat(\n    key_list=[\"驾驶室准乘人数\"],\n    visual_info=visual_info_list,\n    vector_info=vector_info,\n    mllm_predict_info=mllm_predict_info,\n    chat_bot_config=chat_bot_config,\n    retriever_config=retriever_config,\n)\nprint(chat_result)\n```\n\n\u003C\u002Fdetails>\n\n\u003Cdetails>\n   \u003Csummary>\u003Cstrong>4.4 PaddleOCR-VL 示例\u003C\u002Fstrong>\u003C\u002Fsummary>\n\n```python\nfrom paddleocr import PaddleOCRVL\n\npipeline = PaddleOCRVL()\noutput = pipeline.predict(\"https:\u002F\u002Fpaddle-model-ecology.bj.bcebos.com\u002Fpaddlex\u002Fimgs\u002Fdemo_image\u002Fpaddleocr_vl_demo.png\")\nfor res in output:\n    res.print()\n    res.save_to_json(save_path=\"output\")\n    res.save_to_markdown(save_path=\"output\")\n```\n\n\u003C\u002Fdetails>\n\n### 5. 国产异构 AI 加速器\n- [华为昇腾](https:\u002F\u002Fpaddlepaddle.github.io\u002FPaddleOCR\u002Flatest\u002Fversion3.x\u002Fother_devices_support\u002Fpaddlepaddle_install_NPU.html)\n- [昆仑芯](https:\u002F\u002Fpaddlepaddle.github.io\u002FPaddleOCR\u002Flatest\u002Fversion3.x\u002Fother_devices_support\u002Fpaddlepaddle_install_XPU.html)\n\n## 🧩 更多功能\n\n- 将模型转换为 ONNX 格式：[获取 ONNX 模型](https:\u002F\u002Fpaddlepaddle.github.io\u002FPaddleOCR\u002Flatest\u002Fen\u002Fversion3.x\u002Fdeployment\u002Fobtaining_onnx_models.html)。\n- 使用 OpenVINO、ONNX Runtime、TensorRT 等引擎加速推理，或使用 ONNX 格式模型进行推理：[高性能推理](https:\u002F\u002Fpaddlepaddle.github.io\u002FPaddleOCR\u002Flatest\u002Fen\u002Fversion3.x\u002Fdeployment\u002Fhigh_performance_inference.html)。\n- 使用多 GPU 和多进程加速推理：[流水线并行推理](https:\u002F\u002Fpaddlepaddle.github.io\u002FPaddleOCR\u002Flatest\u002Fen\u002Fversion3.x\u002Fpipeline_usage\u002Finstructions\u002Fparallel_inference.html)。\n- 将 PaddleOCR 集成到使用 C++、C#、Java 等编写的应用程序中：[服务化](https:\u002F\u002Fpaddlepaddle.github.io\u002FPaddleOCR\u002Flatest\u002Fen\u002Fversion3.x\u002Fdeployment\u002Fserving.html)。\n\n## ⛰️ 高级教程\n\n- [PP-OCRv5 教程](https:\u002F\u002Fpaddlepaddle.github.io\u002FPaddleOCR\u002Flatest\u002Fversion3.x\u002Fpipeline_usage\u002FOCR.html)\n- [PP-StructureV3 教程](https:\u002F\u002Fpaddlepaddle.github.io\u002FPaddleOCR\u002Flatest\u002Fversion3.x\u002Fpipeline_usage\u002FPP-StructureV3.html)\n- [PP-ChatOCRv4 教程](https:\u002F\u002Fpaddlepaddle.github.io\u002FPaddleOCR\u002Flatest\u002Fversion3.x\u002Fpipeline_usage\u002FPP-ChatOCRv4.html)\n- [PaddleOCR-VL 教程](https:\u002F\u002Fpaddlepaddle.github.io\u002FPaddleOCR\u002Flatest\u002Fversion3.x\u002Fpipeline_usage\u002FPaddleOCR-VL.html)\n\n## 🔄 执行结果快速概览\n\n### PP-OCRv5\n\n\u003Cdiv align=\"center\">\n  \u003Cp>\n       \u003Cimg width=\"100%\" src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FPaddlePaddle_PaddleOCR_readme_95355b710b08.gif\" alt=\"PP-OCRv5 演示\">\n  \u003C\u002Fp>\n\u003C\u002Fdiv>\n\n\n\n### PP-StructureV3\n\n\u003Cdiv align=\"center\">\n  \u003Cp>\n      \u003Cimg width=\"100%\" src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FPaddlePaddle_PaddleOCR_readme_446c93f0ec25.gif\" alt=\"PP-StructureV3 演示\">\n  \u003C\u002Fp>\n\u003C\u002Fdiv>\n\n### PaddleOCR-VL\n\n\u003Cdiv align=\"center\">\n  \u003Cp>\n      \u003Cimg width=\"100%\" src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FPaddlePaddle_PaddleOCR_readme_eba600ad581d.gif\" alt=\"PaddleOCR-VL 演示\">\n  \u003C\u002Fp>\n\u003C\u002Fdiv>\n\n\n## ✨ 敬请期待\n\n⭐ **Star 此仓库以关注激动人心的更新和新版本发布，包括强大的 OCR 和文档解析能力！** ⭐\n\n\u003Cdiv align=\"center\">\n  \u003Cp>\n       \u003Cimg width=\"1200\" src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FPaddlePaddle_PaddleOCR_readme_d8a33a27b69b.gif\" alt=\"Star 项目\">\n  \u003C\u002Fp>\n\u003C\u002Fdiv>\n\n## 👩‍👩‍👧‍👦 社区\n\n\u003Cdiv align=\"center\">\n\n| PaddlePaddle 微信公众号 | 加入技术交流群 |\n| :---: | :---: |\n| \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FPaddlePaddle_PaddleOCR_readme_1bc629737e28.jpg\" width=\"150\"> | \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FPaddlePaddle_PaddleOCR_readme_6e9b6bc2266e.jpg\" width=\"150\"> |\n\u003C\u002Fdiv>\n\n## 😃 利用 PaddleOCR 的精选项目\nPaddleOCR 能有今天的成就离不开其令人惊叹的社区！💗 衷心感谢所有长期合作伙伴、新协作者以及每一位为 PaddleOCR 倾注热情的朋友——无论我们是否公开点名。你们的支持是我们前进的动力！\n\n\u003Cdiv align=\"center\">\n\n| 项目名称 | 描述 |\n| ------------ | ----------- |\n| [RAGFlow](https:\u002F\u002Fgithub.com\u002Finfiniflow\u002Fragflow) \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Finfiniflow\u002Fragflow\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002Finfiniflow\u002Fragflow\">\u003C\u002Fa>|基于深度文档理解的 RAG（检索增强生成）引擎。|\n| [pathway](https:\u002F\u002Fgithub.com\u002Fpathwaycom\u002Fpathway) \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fpathwaycom\u002Fpathway\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002Fpathwaycom\u002Fpathway\">\u003C\u002Fa>|用于流处理、实时分析、LLM（大语言模型）管道和 RAG 的 Python ETL（抽取 - 转换 - 加载）框架。|\n| [MinerU](https:\u002F\u002Fgithub.com\u002Fopendatalab\u002FMinerU) \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fopendatalab\u002FMinerU\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002Fopendatalab\u002FMinerU\">\u003C\u002Fa>|多类型文档转 Markdown 转换工具|\n| [Umi-OCR](https:\u002F\u002Fgithub.com\u002Fhiroi-sora\u002FUmi-OCR) \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fhiroi-sora\u002FUmi-OCR\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002Fhiroi-sora\u002FUmi-OCR\">\u003C\u002Fa>|免费、开源、批量离线 OCR（光学字符识别）软件。|\n| [cherry-studio](https:\u002F\u002Fgithub.com\u002FCherryHQ\u002Fcherry-studio) \u003Ca href=\"https:\u002F\u002Fgithub.com\u002FCherryHQ\u002Fcherry-studio\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002FCherryHQ\u002Fcherry-studio\">\u003C\u002Fa>|支持多个 LLM 提供商的桌面客户端。|\n| [OmniParser](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002FOmniParser)\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002FOmniParser\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002Fmicrosoft\u002FOmniParser\">\u003C\u002Fa> |OmniParser：基于纯视觉的 GUI Agent（图形用户界面智能体）屏幕解析工具。|\n| [QAnything](https:\u002F\u002Fgithub.com\u002Fnetease-youdao\u002FQAnything)\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fnetease-youdao\u002FQAnything\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002Fnetease-youdao\u002FQAnything\">\u003C\u002Fa> |基于任意内容的问答系统。|\n| [PDF-Extract-Kit](https:\u002F\u002Fgithub.com\u002Fopendatalab\u002FPDF-Extract-Kit) \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fopendatalab\u002FPDF-Extract-Kit\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002Fopendatalab\u002FPDF-Extract-Kit\">\u003C\u002Fa>|一个强大的开源工具包，旨在从复杂多样的 PDF 文档中高效提取高质量内容。|\n| [Dango-Translator](https:\u002F\u002Fgithub.com\u002FPantsuDango\u002FDango-Translator)\u003Ca href=\"https:\u002F\u002Fgithub.com\u002FPantsuDango\u002FDango-Translator\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002FPantsuDango\u002FDango-Translator\">\u003C\u002Fa> |识别屏幕上的文本，翻译它并实时显示翻译结果。|\n| [了解更多项目](.\u002Fawesome_projects.md) | [更多基于 PaddleOCR 的项目](.\u002Fawesome_projects.md)|\n\u003C\u002Fdiv>\n\n## 👩‍👩‍👧‍👦 贡献者\n\n\u003Cdiv align=\"center\">\n\u003Ca href=\"https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fgraphs\u002Fcontributors\">\n  \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FPaddlePaddle_PaddleOCR_readme_4c66843d8eb5.png\"  width=\"800\"\u002F>\n\u003C\u002Fa>\n\u003C\u002Fdiv>\n\n## 🌟 星标\n\n\u003Cdiv align=\"center\">\n  \u003Cp>\n      \u003Cimg width=\"800\" src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FPaddlePaddle_PaddleOCR_readme_1606bf1e0578.png\" alt=\"Star-history\">\n  \u003C\u002Fp>\n\u003C\u002Fdiv>\n\n\n## 📄 许可证\n本项目根据 [Apache 2.0 许可证](LICENSE) 发布。\n\n## 🎓 引用\n\n```bibtex\n@misc{cui2025paddleocr30technicalreport,\n      title={PaddleOCR 3.0 Technical Report}, \n      author={Cheng Cui and Ting Sun and Manhui Lin and Tingquan Gao and Yubo Zhang and Jiaxuan Liu and Xueqing Wang and Zelun Zhang and Changda Zhou and Hongen Liu and Yue Zhang and Wenyu Lv and Kui Huang and Yichao Zhang and Jing Zhang and Jun Zhang and Yi Liu and Dianhai Yu and Yanjun Ma},\n      year={2025},\n      eprint={2507.05595},\n      archivePrefix={arXiv},\n      primaryClass={cs.CV},\n      url={https:\u002F\u002Farxiv.org\u002Fabs\u002F2507.05595}, \n}\n\n@misc{cui2025paddleocrvlboostingmultilingualdocument,\n      title={PaddleOCR-VL: Boosting Multilingual Document Parsing via a 0.9B Ultra-Compact Vision-Language Model}, \n      author={Cheng Cui and Ting Sun and Suyin Liang and Tingquan Gao and Zelun Zhang and Jiaxuan Liu and Xueqing Wang and Changda Zhou and Hongen Liu and Manhui Lin and Yue Zhang and Yubo Zhang and Handong Zheng and Jing Zhang and Jun Zhang and Yi Liu and Dianhai Yu and Yanjun Ma},\n      year={2025},\n      eprint={2510.14528},\n      archivePrefix={arXiv},\n      primaryClass={cs.CV},\n      url={https:\u002F\u002Farxiv.org\u002Fabs\u002F2510.14528}, \n}\n```","# PaddleOCR 快速上手指南\n\nPaddleOCR 是一款工业级、生产就绪的 OCR 和文档 AI 引擎，提供从文本提取到智能文档理解的端到端解决方案。支持 100+ 种语言，兼容 CPU、GPU、XPU 及 Ascend NPU 等多种硬件。\n\n## 1. 环境准备\n\n在开始之前，请确保您的开发环境满足以下要求：\n\n- **操作系统**：Linux, Windows, macOS\n- **Python 版本**：3.8 ~ 3.13\n- **硬件支持**：CPU, NVIDIA GPU, Kunlunxin XPU, Ascend NPU\n- **核心依赖**：PaddlePaddle (飞桨深度学习框架)\n\n> 💡 **提示**：推荐使用 NVIDIA GPU 以获得最佳推理性能。若使用国内网络，建议配置镜像源以加速下载。\n\n## 2. 安装步骤\n\n### 2.1 安装 PaddlePaddle\n\n根据您的需求选择 CPU 或 GPU 版本的 PaddlePaddle。国内用户建议使用清华大学开源软件镜像站加速。\n\n```bash\n# 安装 CPU 版本\npip install paddlepaddle -i https:\u002F\u002Fpypi.tuna.tsinghua.edu.cn\u002Fsimple\n\n# 安装 GPU 版本 (需已安装 CUDA 环境)\npip install paddlepaddle-gpu -i https:\u002F\u002Fpypi.tuna.tsinghua.edu.cn\u002Fsimple\n```\n\n### 2.2 安装 PaddleOCR\n\n安装完成后，通过 PyPI 获取 PaddleOCR 包：\n\n```bash\npip install paddleocr -i https:\u002F\u002Fpypi.tuna.tsinghua.edu.cn\u002Fsimple\n```\n\n> ⚠️ **注意**：首次运行时，PaddleOCR 会自动下载默认模型文件（约几百 MB），请确保网络连接稳定。\n\n## 3. 基本使用\n\n### 3.1 Python API 调用\n\n最简单的方式是直接使用 `PaddleOCR` 类进行图像识别。\n\n```python\nfrom paddleocr import PaddleOCR\n\n# 初始化 OCR 对象 (默认加载中英文模型)\nocr = PaddleOCR(use_angle_cls=True, lang=\"ch\")\n\n# 对图片进行识别\nimg_path = 'test_image.jpg'\nresult = ocr.ocr(img_path, cls=True)\n\n# 打印结果\nfor line in result:\n    print(line)\n```\n\n### 3.2 命令行工具\n\n您也可以直接在终端运行预测脚本，无需编写代码：\n\n```bash\n# 基础文字识别\npython tools\u002Finfer\u002Fpredict_system.py --image_dir=.\u002Fdocs\u002Fimgs_en\u002F\n\n# 表格识别 (需要额外安装 tabnet 相关依赖)\npython tools\u002Finfer\u002Fpredict_table.py --image_dir=.\u002Fdocs\u002Fimgs\u002Ftable\u002F\n```\n\n### 3.3 高级功能\n\nPaddleOCR 还支持文档版面分析、公式识别及多模态理解（PaddleOCR-VL）。更多详细用法请参考官方文档或访问 [PaddleOCR 官网](https:\u002F\u002Fwww.paddleocr.com)。","某跨境电商企业的财务团队每周需处理数百份来自不同国家的供应商扫描版发票，用于成本核算与报销审核。\n\n### 没有 PaddleOCR 时\n- 财务人员需手动将图片文字逐字转录到电子表格，耗时巨大且极易出现数字录入错误。\n- 面对混合了英语、日语及特殊符号的复杂发票，通用软件识别率低，常需人工二次校对修正。\n- 提取结果仅为无格式的纯文本，无法直接喂给大模型进行分析，仍需人工整理。\n\n### 使用 PaddleOCR 后\n- PaddleOCR 精准定位表格区域，自动区分金额、日期等关键信息，大幅减少人工干预环节。\n- 凭借百种语言支持能力，无缝处理各国本地化票据，无需针对每种语言切换不同的识别引擎。\n- 直接输出 JSON 等结构化数据，不仅可接入 ERP，还能作为高质量上下文输入给大模型进行智能审核。\n\nPaddleOCR 成功打通了图像文档与数字化流程的壁垒，将原本数小时的人工工作压缩至分钟级完成。","https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FPaddlePaddle_PaddleOCR_f1dc4ea6.png","PaddlePaddle","https:\u002F\u002Foss.gittoolsai.com\u002Favatars\u002FPaddlePaddle_0457ef24.jpg","",null,"http:\u002F\u002Fpaddlepaddle.org","https:\u002F\u002Fgithub.com\u002FPaddlePaddle",[82,86,90,94,98,102,106,109,113,116],{"name":83,"color":84,"percentage":85},"Python","#3572A5",76.9,{"name":87,"color":88,"percentage":89},"C++","#f34b7d",14.1,{"name":91,"color":92,"percentage":93},"Shell","#89e051",5.4,{"name":95,"color":96,"percentage":97},"Java","#b07219",1.2,{"name":99,"color":100,"percentage":101},"Dockerfile","#384d54",0.8,{"name":103,"color":104,"percentage":105},"CMake","#DA3434",0.4,{"name":107,"color":108,"percentage":105},"Cuda","#3A4E3A",{"name":110,"color":111,"percentage":112},"Makefile","#427819",0.3,{"name":114,"color":78,"percentage":115},"Linker Script",0.2,{"name":117,"color":118,"percentage":119},"JavaScript","#f1e05a",0.1,74913,10180,"2026-04-05T10:44:17","Apache-2.0","Linux, Windows, macOS","支持 GPU、NPU、XPU，具体型号及显存要求未说明","未说明",{"notes":128,"python":129,"dependencies":130},"支持 MCP Server 集成；提供 PaddleOCR 3.0 及 VL 技术报告；可通过官方网页在线体验","3.8 ~ 3.13",[131],"paddlepaddle>=3.0",[26,14,13,54],[134,135,136,137,138,139,140,141,142,143,144,145,146],"ocr","chineseocr","pdf2markdown","pp-ocr","pp-structure","document-parsing","document-translation","kie","ai4science","pdf-extractor-rag","pdf-parser","rag","paddleocr-vl",312,"2026-03-27T02:49:30.150509","2026-04-06T01:15:51.879522",[151,156,160,164,168,172],{"id":152,"question_zh":153,"answer_zh":154,"source_url":155},2082,"ONNX 推理时是否需要获取固定的输入形状？","不需要。当前 ONNX 模型已支持动态 shape，因此尝试获取固定的输入 shape 从而进一步处理的操作不再需要。","https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fissues\u002F10223",{"id":157,"question_zh":158,"answer_zh":159,"source_url":155},2083,"PaddleLabel 启动时调用的是环境中的 PaddleOCR 而非本地代码，如何解决？","该问题已验证通过。需确保本地代码路径配置正确，避免调用到全局环境中的 PaddleOCR 包。",{"id":161,"question_zh":162,"answer_zh":163,"source_url":155},2084,"图像处理中遇到通道数不匹配问题怎么办？","该问题已验证通过。建议检查预处理逻辑，确保输入图像的通道数与模型期望的格式一致。",{"id":165,"question_zh":166,"answer_zh":167,"source_url":155},2085,"PyQt 开发中遇到 float 隐式转 int 报错如何处理？","该问题已验证通过。建议在代码中进行显式的类型转换，避免 Python 中 float 隐式转换为 int 导致的潜在错误。",{"id":169,"question_zh":170,"answer_zh":171,"source_url":155},2086,"图像 resize 预处理环节存在什么问题？","相关的图像 resize 预处理问题已验证通过。请参照最新的代码实现或文档规范进行修正，确保预处理流程正确。",{"id":173,"question_zh":174,"answer_zh":175,"source_url":155},2087,"PaddleLabel 自动标注使用的模型是否可以修改？","可以修改。相关功能请求已验证通过，支持用户自定义 PaddleLabel 自动标注时使用的模型。",[177,182,187,192,197,202,207,212,217,222,227,232,237,242,247,252,257,262,267,272],{"id":178,"version":179,"summary_zh":180,"released_at":181},101515,"v3.4.0","## 2026.1.29 v3.4.0 released\r\n* **Release the PaddleOCR-VL-1.5 complex document parsing solution.**\r\n\r\n    PaddleOCR-VL-1.5 is a new iterative version of the PaddleOCR-VL series. Based on comprehensive optimization of the core capabilities of version 1.0, the model achieves **94.5%** accuracy on the authoritative document parsing benchmark OmniDocBench v1.5, surpassing top global general-purpose large models and document parsing–specific models.\r\n    \r\n    PaddleOCR-VL-1.5 innovatively supports **irregular-shaped bounding box localization** of document elements, enabling excellent performance in real-world application scenarios such as **scanning, skew, warping, screen-photography, and complex illumination**, achieving comprehensive SOTA performance. In addition, the model further integrates **seal recognition** and **spotting tasks**, with key metrics continuing to lead mainstream models.\r\n\r\n    You can use it online on the [PaddleOCR official website](https:\u002F\u002Fwww.paddleocr.com) or call the model API.\r\n\r\n* Add support for calling MLX-VLM inference services.\r\n* PaddleOCR-VL now supports cross-page table merging and multi-level heading reconstruction.\r\n* PP-StructureV3 adds support for the `format_block_content` and `markdown_ignore_labels` parameters.\r\n* Fixed an issue where accessing the `\u002Fdocs` endpoint in the official PaddleOCR-VL image would result in an error.\r\n\r\n## 2026.1.29 v3.4.0 发布\r\n* **发布 PaddleOCR-VL-1.5 复杂文档解析方案。**\r\n\r\n    PaddleOCR-VL-1.5 是 PaddleOCR-VL 系列的全新迭代版本。在全面优化 1.0 版本核心能力的基础上，该模型在文档解析权威评测集 OmniDocBench v1.5 上斩获了 **94.5%** 的高精度，超越了全球的顶尖通用大模型及文档解析专用模型。\r\n    \r\n    PaddleOCR-VL-1.5 创新性地支持了文档元素的**异形框定位**，使得 PaddleOCR-VL-1.5 在**扫描、倾斜、弯折、屏幕拍摄及复杂光照**等真实落地场景中均表现卓越，实现了全面的 SOTA。此外，模型进一步集成了**印章识别与文本检测识别任务**，关键指标持续领跑主流模型。\r\n    \r\n    您可以在 [PaddleOCR官网](https:\u002F\u002Fwww.paddleocr.com) 在线使用或者调用该模型的API。\r\n\r\n* 新增对 MLX-VLM 推理服务的调用支持。\r\n* PaddleOCR-VL 支持合并跨页表格、多级标题重建功能。\r\n* PP-StructureV3 支持 `format_block_content`、`markdown_ignore_labels` 参数。\r\n* 修复 PaddleOCR-VL 官方镜像访问 `\u002Fdocs` 接口报错的问题。\r\n\r\n## New Contributors\r\n* @AmirHosseinOmidi0 made their first contribution in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F16659\r\n* @ZhangX-21 made their first contribution in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F16745\r\n* @AdlerFleurant made their first contribution in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F16756\r\n* @tianyuzhou668 made their first contribution in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F16518\r\n* @shiyuasuka made their first contribution in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F17041\r\n* @1250890838 made their first contribution in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F16996\r\n* @Ihebdhouibi made their first contribution in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F16994\r\n* @Ghazi-raad made their first contribution in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F17201\r\n* @orbisai0security made their first contribution in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F17289\r\n* @danghoangnhan made their first contribution in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F17019\r\n* @Luxorion-12 made their first contribution in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F17158\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fcompare\u002Fv3.3.3...v3.4.0","2026-01-29T11:19:45",{"id":183,"version":184,"summary_zh":185,"released_at":186},101516,"v3.3.3","## 2026.1.20 v3.3.3 released\r\n\r\n- PaddleOCR-VL now supports specifying custom model names and API keys, and can seamlessly integrate with inference services from third-party platforms such as SiliconFlow and Novita AI.\r\n- The PP-StructureV3 MCP Server supports using hosted services on the Qianfan platform as the underlying inference engine.\r\n- The documentation for PP-OCRv5 and PaddleOCR-VL has been comprehensively improved, with known errors fixed to enhance readability and accuracy.\r\n- Added support for inference on Muxi GPUs, further expanding hardware compatibility and deployment flexibility.\r\n\r\n## 2026.1.20 v3.3.3 发布\r\n\r\n- PaddleOCR-VL 现已支持指定自定义模型名称与 API Key，并可无缝对接硅基流动、Novita AI 等第三方平台的推理服务。\r\n- PP-StructureV3 MCP Server 支持基于千帆平台的托管服务作为底层推理引擎。\r\n- 全面优化 PP-OCRv5 与 PaddleOCR-VL 相关文档，修复已知错漏，提升可读性与准确性。\r\n- 新增对沐曦 GPU 的推理支持，进一步扩展硬件兼容性与部署灵活性。\r\n\r\n## New Contributors\r\n* @metax666 made their first contribution in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F17269\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fcompare\u002Fv3.3.2...v3.3.3","2026-01-20T07:27:09",{"id":188,"version":189,"summary_zh":190,"released_at":191},101517,"v3.3.2","## 2025.11.13 v3.3.2 released\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fcompare\u002Fv3.3.1...v3.3.2","2025-11-13T14:46:03",{"id":193,"version":194,"summary_zh":195,"released_at":196},101518,"v3.3.1","## 2025.10.29 v3.3.1 released\r\n\r\n- Fixed the issue where the document image preprocessing switch did not take effect in PP-StructureV3 and PaddleOCR-VL.\r\n- Improved the user documentation by adding instructions for using PaddleOCR-VL in a fully offline environment. [Details](https:\u002F\u002Fwww.paddleocr.ai\u002Flatest\u002Fen\u002Fversion3.x\u002Fpipeline_usage\u002FPaddleOCR-VL.html#1-environment-preparation).\r\n\r\n## 2025.10.29 v3.3.1 发布\r\n\r\n- 修复了 PP-StructureV3 和 PaddleOCR-VL 的文档图像预处理开关不生效的问题。\r\n- 优化了使用文档，提供了 PaddleOCR-VL 纯离线环境使用的方式。[详情](https:\u002F\u002Fwww.paddleocr.ai\u002Flatest\u002Fversion3.x\u002Fpipeline_usage\u002FPaddleOCR-VL.html#1)。\r\n\r\n## New Contributors\r\n* @AmirHosseinOmidi0 made their first contribution in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F16659\r\n* @ZhangX-21 made their first contribution in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F16745\r\n* @AdlerFleurant made their first contribution in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F16756\r\n* @tianyuzhou668 made their first contribution in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F16518\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fcompare\u002Fv3.3.0...v3.3.1","2025-10-29T11:49:29",{"id":198,"version":199,"summary_zh":200,"released_at":201},101519,"v3.3.0","## 2025.10.16 v3.3.0 released\r\n\r\n- **Released PaddleOCR-VL:**\r\n    - **Model Introduction:**\r\n        - **PaddleOCR-VL** is a SOTA and resource-efficient model tailored for document parsing. Its core component is PaddleOCR-VL-0.9B, a compact yet powerful vision-language model (VLM) that integrates a NaViT-style dynamic resolution visual encoder with the ERNIE-4.5-0.3B language model to enable accurate element recognition. This innovative model efficiently supports 109 languages and excels in recognizing complex elements (e.g., text, tables, formulas, and charts), while maintaining minimal resource consumption. Through comprehensive evaluations on widely used public benchmarks and in-house benchmarks, PaddleOCR-VL achieves SOTA performance in both page-level document parsing and element-level recognition. It significantly outperforms existing solutions, exhibits strong competitiveness against top-tier VLMs, and delivers fast inference speeds. These strengths make it highly suitable for practical deployment in real-world scenarios. The model has been released on [HuggingFace](https:\u002F\u002Fhuggingface.co\u002FPaddlePaddle\u002FPaddleOCR-VL). Everyone is welcome to download and use it!\r\n\r\n    - **Core Features**:\r\n        - **Compact yet Powerful VLM Architecture**: We present a novel vision-language model that is specifically designed for resource-efficient inference, achieving outstanding performance in element recognition. By integrating a NaViT-style dynamic high-resolution visual encoder with the lightweight ERNIE-4.5-0.3B language model, we significantly enhance the model’s recognition capabilities and decoding efficiency. This integration maintains high accuracy while reducing computational demands, making it well-suited for efficient and practical document processing applications.\r\n        - **SOTA Performance on Document Parsing**: PaddleOCR-VL achieves state-of-the-art performance in both page-level document parsing and element-level recognition. It significantly outperforms existing pipeline-based solutions and exhibiting strong competitiveness against leading vision-language models (VLMs) in document parsing. Moreover, it excels in recognizing complex document elements, such as text, tables, formulas, and charts, making it suitable for a wide range of challenging content types, including handwritten text and historical documents. This makes it highly versatile and suitable for a wide range of document types and scenarios.\r\n        - **Multilingual Support**: PaddleOCR-VL Supports 109 languages, covering major global languages, including but not limited to Chinese, English, Japanese, Latin, and Korean, as well as languages with different scripts and structures, such as Russian (Cyrillic script), Arabic, Hindi (Devanagari script), and Thai. This broad language coverage substantially enhances the applicability of our system to multilingual and globalized document processing scenarios.\r\n\r\n- **Released PP-OCRv5 Multilingual Recognition Model:**\r\n    - Improved the accuracy and coverage of Latin script recognition; added support for Cyrillic, Arabic, Devanagari, Telugu, Tamil, and other language systems, covering recognition of 109 languages. The model has only 2M parameters, and the accuracy of some models has increased by over 40% compared to the previous generation.\r\n\r\n## 2025.10.16 v3.3.0 发布\r\n\r\n- **发布PaddleOCR-VL**：\r\n    - **模型介绍：**\r\n        - **PaddleOCR-VL** 是一款先进、高效的文档解析模型，专为文档中的元素识别设计。其核心组件为 PaddleOCR-VL-0.9B，这是一种紧凑而强大的视觉语言模型（VLM），它由 NaViT 风格的动态分辨率视觉编码器与 ERNIE-4.5-0.3B 语言模型组成，能够实现精准的元素识别。该模型支持 109 种语言，并在识别复杂元素（如文本、表格、公式和图表）方面表现出色，同时保持极低的资源消耗。通过在广泛使用的公开基准与内部基准上的全面评测，PaddleOCR-VL 在页级级文档解析与元素级识别均达到 SOTA 表现。它显著优于现有的基于Pipeline方案和文档解析多模态方案以及先进的通用多模态大模型，并具备更快的推理速度。这些优势使其非常适合在真实场景中落地部署。模型已发布至[HuggingFace](https:\u002F\u002Fhuggingface.co\u002FPaddlePaddle\u002FPaddleOCR-VL)，欢迎大家下载使用！\r\n\r\n    - **特性：**\r\n        - **紧凑而强大的视觉语言模型架构**：我们提出了一种新的视觉语言模型，专为资源高效的推理而设计，在元素识别方面表现出色。通过将NaViT风格的动态高分辨率视觉编码器与轻量级的ERNIE-4.5-0.3B语言模型结合，我们显著增强了模型的识别能力和解码效率。这种集成在保持高准确率的同时降低了计算需求，使其非常适合高效且实用的文档处理应用。\r\n        - **文档解析的SOTA性能**：PaddleOCR-VL在页面级文档解析和元素级识别中达到了最先进的性能。它显著优于现有的基于流水线的解决方案，并在文档解析中展现出与领先的视觉语言模型（VLMs）竞争的强劲实力。此外，它在识别复杂的文档元素（如文本、表格、公式和图表）方面表现出色，使其适用于包括手写文本和历史文献在内的各种具有挑战性的内容类型。这使得它具有高度的多功能性，适用于广泛的文档类型和场景。\r\n        - **多语言支持**：PaddleOCR-VL支持109种语言，覆盖了主要的全球语言，包括但不限于中文、英文、日文、拉丁文和韩文，以及使用不同文字和结构的语言，如俄语（西里尔字母）、阿拉伯语、印地语（天城文）和泰语。这种广泛的语言覆盖大大增强了我们系统在多语言和全球化文档处理场景中的适用性。\r\n\r\n- **发布PP-OCRv5小语种识别模型**：\r\n    - 优化拉丁文识别的准度和广度，新增西里尔文、阿拉伯文、天城文、泰卢固语、泰米尔语等语系，覆盖109种语言文字的识别。模型参数量仅为2M，部分模型精度较上一代提升40%以上。","2025-10-16T12:58:29",{"id":203,"version":204,"summary_zh":205,"released_at":206},101520,"v3.2.0","## 2025.8.21 v3.2.0 released\r\n\r\n- **Significant Model Additions:**\r\n    - Introduced training, inference, and deployment for PP-OCRv5 recognition models in English, Thai, and Greek. **The PP-OCRv5 English model delivers an 11% improvement in English scenarios compared to the main PP-OCRv5 model, with the Thai and Greek recognition models achieving accuracies of 82.68% and 89.28%, respectively.**\r\n\r\n- **Deployment Capability Upgrades:**\r\n    - **Full support for PaddlePaddle framework versions 3.1.0 and 3.1.1.**\r\n    - **Comprehensive upgrade of the PP-OCRv5 C++ local deployment solution, now supporting both Linux and Windows, with feature parity and identical accuracy to the Python implementation.**\r\n    - **High-performance inference now supports CUDA 12, and inference can be performed using either the Paddle Inference or ONNX Runtime backends.**\r\n    - **The high-stability service-oriented deployment solution is now fully open-sourced, allowing users to customize Docker images and SDKs as required.**\r\n    - The high-stability service-oriented deployment solution also supports invocation via manually constructed HTTP requests, enabling client-side code development in any programming language.\r\n\r\n- **Benchmark Support:**\r\n    - **All production lines now support fine-grained benchmarking, enabling measurement of end-to-end inference time as well as per-layer and per-module latency data to assist with performance analysis.**\r\n    - **Documentation has been updated to include key metrics for commonly used configurations on mainstream hardware, such as inference latency and memory usage, providing deployment references for users.**\r\n\r\n- **Bug Fixes:**\r\n    - Resolved the issue of failed log saving during model training.\r\n    - Upgraded the data augmentation component for formula models for compatibility with newer versions of the albumentations dependency, and fixed deadlock warnings when using the tokenizers package in multi-process scenarios.\r\n    - Fixed inconsistencies in switch behaviors (e.g., `use_chart_parsing`) in the PP-StructureV3 configuration files compared to other pipelines.\r\n\r\n- **Other Enhancements:**\r\n    - **Separated core and optional dependencies. Only minimal core dependencies are required for basic text recognition; additional dependencies for document parsing and information extraction can be installed as needed.**\r\n    - **Enabled support for NVIDIA RTX 50 series graphics cards on Windows; users can refer to the installation guide for the corresponding PaddlePaddle framework versions.**\r\n    - **PP-OCR series models now support returning single-character coordinates.**\r\n    - Added AIStudio, ModelScope, and other model download sources, allowing users to specify the source for model downloads.\r\n    - Added support for chart-to-table conversion via the PP-Chart2Table module.\r\n    - Optimized documentation descriptions to improve usability.\r\n\r\n## 2025.8.21 v3.2.0 发布\r\n\r\n- **重要模型新增：**\r\n    - 新增 PP-OCRv5 英文、泰文、希腊文识别模型的训练、推理、部署。**其中 PP-OCRv5 英文模型较 PP-OCRv5 主模型在英文场景提升 11%，泰文识别模型精度 82.68%，希腊文识别模型精度 89.28%。**\r\n\r\n- **部署能力升级：**\r\n    - **全面支持飞桨框架 3.1.0 和 3.1.1 版本。**\r\n    - **全面升级 PP-OCRv5 C++ 本地部署方案，支持 Linux、Windows，功能及精度效果与 Python 方案保持一致。**\r\n    - **高性能推理支持 CUDA 12，可使用 Paddle Inference、ONNX Runtime 后端推理。**\r\n    - **高稳定性服务化部署方案全面开源，支持用户根据需求对 Docker 镜像和 SDK 进行定制化修改。**\r\n    - 高稳定性服务化部署方案支持通过手动构造HTTP请求的方式调用，该方式允许客户端代码使用任意编程语言编写。\r\n\r\n- **Benchmark支持**：\r\n    - **全部产线支持产线细粒度 benchmark，能够测量产线端到端推理时间以及逐层、逐模块的耗时数据，可用于辅助产线性能分析。**\r\n    - **文档中补充各产线常用配置在主流硬件上的关键指标，包括推理耗时和内存占用等，为用户部署提供参考。**\r\n\r\n- **Bug修复：**\r\n    - 修复模型训练时训练日志保存失败的问题。\r\n    - 对公式模型的数据增强部分进行了版本兼容性升级，以适应新版本的 albumentations 依赖，并修复了在多进程使用 tokenizers 依赖包时出现的死锁警告。\r\n    - 修复 PP-StructureV3 配置文件中的 `use_chart_parsing` 等开关行为与其他产线不统一的问题。\r\n\r\n- **其他升级：**\r\n    - **分离必要依赖与可选依赖。使用基础文字识别功能时，仅需安装少量核心依赖；若需文档解析、信息抽取等功能，用户可按需选择安装额外依赖。**\r\n    - **支持 Windows 用户使用英伟达 50 系显卡，可根据安装文档安装对应版本的 paddle 框架。**\r\n    - **PP-OCR 系列模型支持返回单文字坐标。**\r\n    - 模型新增 AIStudio、ModelScope 等下载源。可指定相关下载源下载对应的模型。\r\n    - 支持图表转表 PP-Chart2Table 单功能模块推理能力。\r\n    - 优化部分使用文档中的描述，提升易用性。\r\n    \r\n## New Contributors\r\n* @awkrail made their first contribution in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F16262\r\n* @liuwake made their first contribution in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F16281\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fcompare\u002Fv3.1.1...v3.2.0","2025-08-21T11:11:07",{"id":208,"version":209,"summary_zh":210,"released_at":211},101521,"v3.1.1","## 2025.8.15 v3.1.1 released\r\n\r\n- **Bug Fixes:**\r\n  - Added the missing methods `save_vector`, `save_visual_info_list`, `load_vector`, and `load_visual_info_list` in the `PP-ChatOCRv4` class.\r\n  - Added the missing parameters `glossary` and `llm_request_interval` to the `translate` method in the `PPDocTranslation` class.\r\n\r\n- **Documentation Improvements:**\r\n  - Added a demo to the MCP documentation.\r\n  - Added information about the PaddlePaddle and PaddleOCR version used for performance metrics testing in the documentation.\r\n  - Fixed errors and omissions in the production line document translation.\r\n\r\n- **Others:**\r\n  - Changed the MCP server dependency to use the pure Python library `puremagic` instead of `python-magic` to reduce installation issues.\r\n  - Retested PP-OCRv5 performance metrics with PaddleOCR version 3.1.0 and updated the documentation.\r\n\r\n## 2025.8.15 v3.1.1 发布\r\n- **bug修复：**\r\n  - 补充 `PP-ChatOCRv4` 类缺失的`save_vector`、`save_visual_info_list`、`load_vector、load_visual_info_list` 方法。\r\n  - 补充 `PPDocTranslation` 类的 `translate` 方法缺失的 `glossary 和 `llm_request_interval 参数。\r\n\r\n- **文档优化：**\r\n  - 补充 MCP 文档中的 demo。\r\n  - 补充文档中测试性能指标使用的飞桨框架与 PaddleOCR 版本。\r\n  - 修复文档翻译产线文档中的错漏。\r\n\r\n- **其他：**\r\n  - 修改 MCP 服务器依赖，使用纯 Python 库 `puremagic` 代替 `python-magic`，减少安装问题。\r\n  - 使用 3.1.0 版本 PaddleOCR 重新测试 PP-OCRv5 性能指标，更新文档。\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fcompare\u002Fv3.1.0...v3.1.1","2025-08-15T08:55:16",{"id":213,"version":214,"summary_zh":215,"released_at":216},101522,"v3.1.0","## 2025.6.29 v3.1.0 released\r\n\r\n- **Key Models and Pipelines:**\r\n  - **Added PP-OCRv5 Multilingual Text Recognition Model**, which supports the training and inference process for text recognition models in 37 languages, including French, Spanish, Portuguese, Russian, Korean, etc. **Average accuracy improved by over 30%.** [Details](https:\u002F\u002Fpaddlepaddle.github.io\u002FPaddleOCR\u002Flatest\u002Fen\u002Fversion3.x\u002Falgorithm\u002FPP-OCRv5\u002FPP-OCRv5_multi_languages.html)\r\n  - Upgraded the **PP-Chart2Table model** in PP-StructureV3, further enhancing the capability of converting charts to tables. On internal custom evaluation sets, the metric (RMS-F1) **increased by 9.36 percentage points (71.24% -> 80.60%).**\r\n  - Newly launched **document translation pipeline, PP-DocTranslation, based on PP-StructureV3 and ERNIE 4.5 Turbo**, which supports the translation of Markdown format documents, various complex-layout PDF documents, and document images, with the results saved as Markdown format documents. [Details](https:\u002F\u002Fpaddlepaddle.github.io\u002FPaddleOCR\u002Flatest\u002Fen\u002Fversion3.x\u002Fpipeline_usage\u002FPP-DocTranslation.html)\r\n\r\n\r\n- **New MCP server:** [Details](https:\u002F\u002Fpaddlepaddle.github.io\u002FPaddleOCR\u002Flatest\u002Fen\u002Fversion3.x\u002Fdeployment\u002Fmcp_server.html)\r\n  - **Supports both OCR and PP-StructureV3 pipelines.**\r\n  - Supports three working modes: local Python library, AIStudio Community Cloud Service, and self-hosted service.\r\n  - Supports invoking local services via stdio and remote services via Streamable HTTP.\r\n\r\n- **Documentation Optimization:** Improved the descriptions in some user guides for a smoother reading experience.\r\n\r\n## 2025.6.29 v3.1.0 发布\r\n\r\n- **重要模型和产线：**\r\n  - **新增 PP-OCRv5 多语种文本识别模型**，支持法语、西班牙语、葡萄牙语、俄语、韩语等 37 种语言的文字识别模型的训推流程。**平均精度涨幅超30%。**[详情](https:\u002F\u002Fpaddlepaddle.github.io\u002FPaddleOCR\u002Flatest\u002Fversion3.x\u002Falgorithm\u002FPP-OCRv5\u002FPP-OCRv5_multi_languages.html)\r\n  - 升级 PP-StructureV3 中的 **PP-Chart2Table 模型**，图表转表能力进一步升级，在内部自建测评集合上指标（RMS-F1）**提升 9.36 个百分点（71.24% -> 80.60%）。**\r\n  - 新增基于 PP-StructureV3 和 ERNIE 4.5 Turbo 的**文档翻译产线 PP-DocTranslation，支持翻译 Markdown 格式文档、各种复杂版式的 PDF 文档和文档图像，结果保存为 Markdown 格式文档。**[详情](https:\u002F\u002Fpaddlepaddle.github.io\u002FPaddleOCR\u002Flatest\u002Fversion3.x\u002Fpipeline_usage\u002FPP-DocTranslation.html)\r\n\r\n- **新增MCP server：**[详情](https:\u002F\u002Fpaddlepaddle.github.io\u002FPaddleOCR\u002Flatest\u002Fversion3.x\u002Fdeployment\u002Fmcp_server.html)\r\n  - **支持 OCR 和 PP-StructureV3 两种工具；**\r\n  - 支持本地Python库、星河社区云服务、自托管服务三种工作模式；\r\n  - 支持通过 stdio 调用本地服务，通过 Streamable HTTP 调用远程服务。\r\n\r\n- **文档优化：** 优化了部分使用文档描述，提升阅读体验。\r\n\r\n\r\n\r\n\r\n\r\n","2025-06-29T06:57:35",{"id":218,"version":219,"summary_zh":220,"released_at":221},101523,"v3.0.3","- **Bug修复：** \r\n  - 修复`enable_mkldnn`参数不生效的问题，恢复CPU默认使用MKL-DNN推理的行为。\r\n  - [随PaddleX 3.0.3 版本的其他修复](https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleX\u002Freleases\u002Ftag\u002Fv3.0.3)","2025-06-26T10:04:31",{"id":223,"version":224,"summary_zh":225,"released_at":226},101524,"v3.0.2","- **功能新增：**\r\n  - 模型默认下载源从`BOS`改为`HuggingFace`，同时也支持用户通过更改环境变量`PADDLE_PDX_MODEL_SOURCE`为`BOS`，将模型下载源设置为百度云对象存储BOS。\r\n  - PP-OCRv5、PP-StructureV3、PP-ChatOCRv4等pipeline新增C++、Java、Go、C#、Node.js、PHP 6种语言的服务调用示例。\r\n  - 优化PP-StructureV3产线中版面分区排序算法，对复杂竖版版面排序逻辑进行完善，进一步提升了复杂版面排序效果。\r\n  - 优化模型选择逻辑，当指定语言、未指定模型版本时，自动选择支持该语言的最新版本的模型。 \r\n  -  为MKL-DNN缓存大小设置默认上界，防止缓存无限增长。同时，支持用户配置缓存容量。\r\n  - 更新高性能推理默认配置，支持Paddle MKL-DNN加速。优化高性能推理自动配置逻辑，支持更智能的配置选择。\r\n  - 调整默认设备获取逻辑，考虑环境中安装的Paddle框架对计算设备的实际支持情况，使程序行为更符合直觉。\r\n  - 新增PP-OCRv5的Android端示例，[详情](https:\u002F\u002Fpaddlepaddle.github.io\u002FPaddleOCR\u002Flatest\u002Fversion3.x\u002Fdeployment\u002Fon_device_deployment.html)。\r\n\r\n- **Bug修复：**\r\n  - 修复PP-StructureV3部分CLI参数不生效的问题。\r\n  - 修复部分情况下`export_paddlex_config_to_yaml`无法正常工作的问题。\r\n  - 修复save_path实际行为与文档描述不符的问题。\r\n  - 修复基础服务化部署在使用MKL-DNN时可能出现的多线程错误。\r\n  - 修复Latex-OCR模型的图像预处理的通道顺序错误。\r\n  - 修复文本识别模块保存可视化图像的通道顺序错误。\r\n  - 修复PP-StructureV3中表格可视化结果通道顺序错误。\r\n  - 修复PP-StructureV3产线中极特殊的情况下，计算overlap_ratio时，变量溢出问题。\r\n\r\n- **文档优化：**\r\n  - 更新文档中对`enable_mkldnn`参数的说明，使其更准确地描述程序的实际行为。\r\n  - 修复文档中对`lang`和`ocr_version`参数描述的错误。\r\n  - 补充通过CLI导出产线配置文件的说明。\r\n  - 修复PP-OCRv5性能数据表格中的列缺失问题。\r\n  - 润色PP-StructureV3在不同配置下的benchmark指标。\r\n\r\n- **其他：**\r\n  - 放松numpy、pandas等依赖的版本限制，恢复对Python 3.12的支持。","2025-06-18T16:38:08",{"id":228,"version":229,"summary_zh":230,"released_at":231},101525,"v3.0.1","- **优化部分模型和模型配置：**\r\n  - 更新 PP-OCRv5默认模型配置，检测和识别均由mobile改为server模型。为了改善大多数的场景默认效果，配置中的参数`limit_side_len`由736改为64\r\n  - 新增文本行方向分类`PP-LCNet_x1_0_textline_ori`模型，精度99.42%，OCR、PP-StructureV3、PP-ChatOCRv4产线的默认文本行方向分类器改为该模型\r\n  - 优化文本行方向分类`PP-LCNet_x0_25_textline_ori`模型，精度提升3.3个百分点，当前精度98.85%\r\n- **优化3.0.0版本部分存在的问题**\r\n  - **优化CLI使用体验：** 当使用PaddleOCR CLI不传入任何参数时，给出用法提示。\r\n  - **新增参数：** PP-ChatOCRv3、PP-StructureV3支持`use_textline_orientation`参数。\r\n  - **CPU推理速度优化：** 所有产线CPU推理默认开启MKL-DNN。\r\n  - **C++推理支持：** PP-OCRv5的检测和识别串联部分支持C++推理\r\n- **修复3.0.0版本部分存在的问题**\r\n  - 修复由于公式识别、表格识别模型无法使用MKL-DNN导致PP-StructureV3在部分cpu推理报错的问题\r\n  - 修复在部分GPU环境中推理报`FatalError: Process abort signal is detected by the operating system`错误的问题\r\n  - 修复部分Python3.8环境的type hint的问题\r\n  - 修复`PPStructureV3.concatenate_markdown_pages`方法不存在的问题。\r\n  - 修复实例化`paddleocr.PaddleOCR`时同时指定`lang`和`model_name`时`model_name`不生效的问题。","2025-06-05T03:27:00",{"id":233,"version":234,"summary_zh":235,"released_at":236},101526,"v3.0.0","- **发布全场景文字识别模型PP-OCRv5：** 单模型支持五种文字类型和复杂手写体识别；整体识别精度相比上一代提升13个百分点。\r\n \r\n- **发布通用文档解析方案PP-StructureV3：** 支持多场景、多版式 PDF 高精度解析，在公开评测集中领先众多开源和闭源方案。\r\n\r\n- **发布智能文档理解方案PP-ChatOCRv4：** 原生支持文心大模型4.5 Turbo，精度相比上一代提升15个百分点。\r\n\r\n- **重构部署能力，统一推理接口：** PaddleOCR 3.0 融合了飞桨 PaddleX3.0 工具的底层能力，全面升级推理、部署模块，优化 2.x 版本的设计，统一并优化了 Python API 和命令行接口（CLI）。部署能力现覆盖高性能推理、服务化部署及端侧部署三大场景。\r\n\r\n- **适配飞桨框架 3.0，优化训练流程：** 新版本已兼容飞桨 3.0 的 CINN 编译器等最新特性，静态图模型存储文件名由 `xxx.pdmodel` 改为 `xxx.json`。\r\n\r\n- **统一模型名称：** 对PaddleOCR3.0支持的模型命名体系进行了更新，采用更规范、统一的命名规则，为后续迭代与维护奠定基础。","2025-05-20T12:16:51",{"id":238,"version":239,"summary_zh":240,"released_at":241},101527,"v2.10.0","## What's Changed\r\n* update docs by @cuicheng01 in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F14031\r\n* update paddle2onnx doc by @inisis in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F14038\r\n* fix gpu memory growth by @zhangyubo0722 in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F14037\r\n* updata en docs by @dyning in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F14036\r\n* fix nan in PP-OCRv4 by @wangna11BD in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F14043\r\n* update a live promotion by @Zhiiixin in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F14042\r\n* reset latex ocr by @zhangyubo0722 in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F14046\r\n* Update pyproject.toml for add dependency by @Liyulingyue in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F14058\r\n* Fix `CMAKE_CXX_FLAGS` optimize flag by @Hirozy in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F14059\r\n* fix isnan_v2 is not supported in paddle2onnx by @GreatV in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F14060\r\n* ci: Fixed docs multi version error by @SWHL in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F14048\r\n* fix hyperlinks by @AmberC0209 in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F14073\r\n* fix nan in ppocrv4 for benchmark by @wangna11BD in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F14072\r\n* ci: Support seperate update of branch docs by @SWHL in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F14079\r\n* ci: fixed main doc ci by @SWHL in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F14084\r\n* Allow `create_predictor` function to accept array of ONNX Execution Providers by @Salmondx in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F14078\r\n* docs: update quickstart by @SWHL in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F14108\r\n* docs: add command line usage documentation of quickstart page by @SWHL in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F14110\r\n* docs: add installation documentation of paddle by @SWHL in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F14117\r\n* docs: fixed typo by @SWHL in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F14118\r\n* image without any text will show a warning by @GreatV in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F14132\r\n* doc: remove duplicate paragraphs by @GreatV in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F14133\r\n* docs: update paddle2onnx documentations by @GreatV in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F14144\r\n* [third-party] Fix the issue of inference errors with KIE mode in ONNX format by @Alex37882388 in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F14138\r\n* update tests PR CI github action by @GreatV in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F14159\r\n* 移除doc目录下文档，保留fonts和doc_i18n两个目录 by @SWHL in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F14156\r\n* 移除ppstructure目录下旧有文档 by @SWHL in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F14161\r\n* docs: fixed error image link (#14164) by @SWHL in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F14165\r\n* 更新i18n的首页内容到新站点 by @SWHL in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F14166\r\n* docs: fix i18n languange code error by @SWHL in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F14167\r\n* docs: fix syntax error by @SWHL in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F14168\r\n* docs: update i18n docs by @GreatV in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F14169\r\n* upgrade to numpy 2.0 and remove imgaug by @GreatV in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F13937\r\n* docs: format multi languange docs home page by @SWHL in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F14170\r\n* docs: add the missing image by @GreatV in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F14180\r\n* Create close_inactive_issues.yaml by @GreatV in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F14183\r\n* update hpi config by @zhangyubo0722 in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F14076\r\n* Update close_inactive_issues.yaml by @GreatV in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F14189\r\n* Update close_inactive_issues.yaml by @GreatV in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F14190\r\n* remove lock inactive issues by @GreatV in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F14192\r\n* fix benchmark bug by @changdazhou in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F14194\r\n* pre-commit autoupdate && pre-commit run --all-files by @cclauss in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F14201\r\n* Remove Python 2 compatibility dependency six by @cclauss in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F14202\r\n* update quick_start by @AmberC0209 in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F14200\r\n* rename train result by @zhangyubo0722 in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F14217\r\n* fix benchmark bug by @changdazhou in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F14235\r\n* fix benchmark det_r50_vd_pse_v2_0 train error by @GreatV in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F14239\r\n* update infer\u002Futility.py to support json format model by @GreatV in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F14233\r","2025-03-07T07:03:56",{"id":243,"version":244,"summary_zh":245,"released_at":246},101528,"v2.9.1","## What's Changed\r\n* [cherry-pick] update paddle2onnx doc by @inisis in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F14051\r\n* [cherry-pick] fix isnan is not supported and update pyproject by @GreatV in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F14061\r\n* [cherry-pick] ci: Fixed docs multi version error by @GreatV in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F14063\r\n\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fcompare\u002Fv2.9.0...v2.9.1","2024-10-22T05:57:17",{"id":248,"version":249,"summary_zh":250,"released_at":251},101529,"v2.9.0","## What's Changed\r\n* fix: table recognition content is not escaped properly by @GreatV in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F13277\r\n* fix bug when layout_predictor is None by @GreatV in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F13279\r\n* add url in pyproject, and update version number by @GreatV in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F13274\r\n* unifying data types in the SLAHead by @GreatV in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F13276\r\n* add PaddleX info to README by @TingquanGao in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F13308\r\n* Update expired link in quickstart.md by @ZeddYu in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F13253\r\n* optimize func: get_infer_gpuid by @GreatV in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F13275\r\n* fix slice op parameters not being passed correctly by @GreatV in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F13319\r\n* Solve ModuleNotFoundError: No module named 'tools.infer' by @myhloli in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F13348\r\n* Add hardware docs by @nepeplwu in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F13329\r\n* add paddlex link by @TingquanGao in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F13316\r\n* Fix the dictionary bug in tablerec inference by @Topdu in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F13362\r\n* add bn_dict.txt by @taeefnajib in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F13373\r\n* add missing docstring in paddleocr.py using copilot by @jzhang533 in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F13344\r\n* line 445 program.py by @ManikSinghSarmaal in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F13389\r\n* fix layout recovery import error by @GreatV in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F13434\r\n* Latexocr paddle by @liuhongen1234567 in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F13401\r\n* [doc]add amp train notes for detection train by @andyjiang1116 in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F13481\r\n* remove some of the less common dependencies by @GreatV in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F13461\r\n* docs: Add a new document site by @SWHL in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F13375\r\n* Update mkdocs.yml by @GreatV in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F13487\r\n* chore: Update issue template by @SWHL in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F13505\r\n* chore: Update bug report template by @SWHL in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F13508\r\n* Fix cpp_infer \"--enable_mkldnn=false\" not effective by @hiroi-sora in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F13539\r\n* Enable Main Branch Support for PaddleX by @zhangyubo0722 in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F13523\r\n* docs: Update README by @SWHL in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F13543\r\n* docs: Update README_en by @SWHL in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F13545\r\n* 修改错别字 by @MonkeyBrothers in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F13544\r\n* docs: Remove old applications docs by @SWHL in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F13551\r\n* fix: 'numpy' has no attribute 'astype' by @laolitou in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F13554\r\n* add latexocr docs and fix some typos by @GreatV in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F13532\r\n* chore(Issue_template): Add validation of Environment and MPE code by @SWHL in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F13559\r\n* skip text files when running test ci by @GreatV in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F13561\r\n* fix bug for paddlepaddle3.0 by @changdazhou in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F13568\r\n* docs: Update the pdf file path in the operation demonstration by @Gmgge in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F13575\r\n* support benchmark for paddlepaddle3.0 by @changdazhou in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F13574\r\n* improve the reading experience of some documents by @GreatV in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F13562\r\n* update dive into OCR book link by @GreatV in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F13581\r\n* docs: Shorten the image path and remove dupliate images by @SWHL in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F13585\r\n* docs: Fix docs errors by @SWHL in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F13588\r\n* skip text files when running test ci on push by @GreatV in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F13582\r\n* docs: Add android_demo docs by @SWHL in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F13601\r\n* fix download bug when use multi gpus by @changdazhou in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F13610\r\n* disable automatic checks for new version albumentations by @GreatV in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F13583\r\n* 修复LaTeXOCR 在paddleX中的一些问题 by @liuhongen1234567 in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F13646\r\n* update docs and remove out-of-date event by @GreatV in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F13660\r\n* setuptools 72.2.0 ","2024-10-18T15:43:04",{"id":253,"version":254,"summary_zh":255,"released_at":256},101530,"v2.8.1","## What's Changed\r\n* [cherry-pick] add project url and fix a bug by @GreatV in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F13281\r\n* [cherry-pick] fix slice op parameters not being passed correctly (#13319) by @GreatV in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F13324\r\n* Fix the dictionary bug in tablerec inference by @Topdu in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F13364\r\n\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fcompare\u002Fv2.8.0...v2.8.1","2024-07-17T10:48:47",{"id":258,"version":259,"summary_zh":260,"released_at":261},101531,"v2.8.0","**[终于等到你！PaddleOCR 新版本发布！](https:\u002F\u002Fpfcc.blog\u002Fposts\u002Fpaddleocr-release)**\r\n\r\n## What's Changed\r\n* [Cherry-pick] #10515 by @ToddBear in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F10537\r\n* [BugFix]compat_pillow by @shiyutang in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F10596\r\n* [bug fix] fix none res in recovery by @andyjiang1116 in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F10603\r\n* Fix seed passing issue of build_dataloader by @RuohengMa in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F10614\r\n* [bug fix]rm invalid params by @andyjiang1116 in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F10605\r\n* [Cherry-pick] #10441 #10512 by @moehuster in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F10593\r\n* 修改数据增强导致的DSR报错 by @xu-peng-7 in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F10662\r\n* onnxruntime support gpu by @WenmuZhou in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F10668\r\n* Update VQA to use the updated LayoutLM syntax from PaddleNLP by @sijunhe in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F9791\r\n* 实现功能：当--savefile为true时，在--output下以当前图片名称后接“.txt”为文件名保存ocr推理结果，解决了issues： by @WilliamQf-AI in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F10628\r\n* Cherrypicking GH-10217 and GH-10216 to PaddlePaddle:dygraph by @UserUnknownFactor in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F10654\r\n* fix numpy speed by @wanghuancoder in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F10773\r\n* Cherrypicking GH-10251 & GH-10181 to PaddleOCR:dygraph by @itasli in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F10710\r\n* rec_r45_abinet.yml add max_length and image_size by @xlg-go in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F10744\r\n* ch_PP-OCRv4_rec_distill.yml, fix KeyError: 'NRTRLabelDecode' by @xlg-go in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F10761\r\n* 根据推理对三通道的图像需求，以及opencv中imread参数说明IMREAD_COLOR(If set, always convert … by @Gmgge in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F10777\r\n* Update algorithm_kie_vi_layoutxlm_en.md by @sagarjgb in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F10736\r\n* Add new recognition method \"ParseQ\" by @ToddBear in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F10836\r\n* rm fluid for paddle dev by @tink2123 in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F10931\r\n* rec_r45_abinet for export model by @xlg-go in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F10892\r\n* fix:修复通道数不匹配造成的PPOCRLabel启动失败问题#10748,根据更新日志发现#10655，由于paddleocr中增加了对… by @Gmgge in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F10847\r\n* [New] add rec CPPD model by @Topdu in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F10990\r\n* fix `cls_x` and `bbox_x` is possibly unbound by @SigureMo in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F10991\r\n* add svtr large model by @zhangyubo0722 in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F10937\r\n* [WIP]support eval pre epoch by @zhangyubo0722 in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F11003\r\n* Update kie_datasets_en.md by @sagarjgb in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F10735\r\n* fix import collection for py310 by @tink2123 in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F11012\r\n* update ppocrv4_framework by @tink2123 in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F11048\r\n* Update how_to_do_kie_en.md by @sagarjgb in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F10731\r\n* add cppd u14m train model and doc by @Topdu in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F11052\r\n* Fixed bug with \"max_text_length\" for VisionLAN by @victor30608 in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F11025\r\n* Cherrypicking GH-10923 to PaddleOCR:dygraph by @itasli in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F11069\r\n* Update quickstart_en.md by @sagarjgb in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F10732\r\n* Update README.md by @sagarjgb in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F10733\r\n* Update algorithm_overview_en.md by @sagarjgb in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F10734\r\n* [Cherry-pick] Cherry-pick from release\u002F2.6 by @shiyutang in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F11092\r\n* [TIPC]update tipc scripts by @USTCKAY in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F11097\r\n* fix satrn export for paddle2.5 by @tink2123 in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F11096\r\n* [BugFix]Fix parseq net by @shiyutang in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F11126\r\n* update uygur dict by @hfengzhi in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F11125\r\n* Add tipc for \"ParseQ\" method by @ToddBear in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F10843\r\n* fix SAR inference, when batch size>1, norm_img_batch and valid_ratios… by @shiyunalex in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F11238\r\n* v4 det cml configs by @sylarwcy in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F11258\r\n* 解決recognition的train test分割程式執行後的文檔每行間多出一行空格 by @DingHsun in https:\u002F\u002Fgithub.com\u002FPaddlePaddle\u002FPaddleOCR\u002Fpull\u002F11280\r\n* Fix for Ambiguous Boolean Evaluation Error in PaddleOCR with Python 3.11 by @muhammadAgfian96 in https:\u002F\u002Fgithub.com\u002F","2024-07-04T11:45:32",{"id":263,"version":264,"summary_zh":265,"released_at":266},101532,"v2.7.5","fix broken v2.7.4","2024-03-29T09:48:32",{"id":268,"version":269,"summary_zh":270,"released_at":271},101533,"v2.7.4","This release contains the missed commits from v2.7.0 to v2.7.1.\r\nfixed : #11824","2024-03-29T02:47:10",{"id":273,"version":274,"summary_zh":275,"released_at":276},101534,"v2.7.3","## What's Changed\r\nfixed #11808","2024-03-28T03:46:32"]