[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"similar-matlab-deep-learning--MATLAB-Deep-Learning-Model-Hub":3,"tool-matlab-deep-learning--MATLAB-Deep-Learning-Model-Hub":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":70,"readme_en":71,"readme_zh":72,"quickstart_zh":73,"use_case_zh":74,"hero_image_url":75,"owner_login":76,"owner_name":77,"owner_avatar_url":78,"owner_bio":79,"owner_company":80,"owner_location":80,"owner_email":80,"owner_twitter":80,"owner_website":81,"owner_url":82,"languages":83,"stars":88,"forks":89,"last_commit_at":90,"license":91,"difficulty_score":23,"env_os":92,"env_gpu":93,"env_ram":94,"env_deps":95,"category_tags":103,"github_topics":104,"view_count":23,"oss_zip_url":80,"oss_zip_packed_at":80,"status":16,"created_at":111,"updated_at":112,"faqs":113,"releases":128},3140,"matlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub","MATLAB-Deep-Learning-Model-Hub","Discover pretrained models for deep learning in MATLAB","MATLAB Deep Learning Model Hub 是一个专为 MATLAB 用户打造的预训练深度学习模型资源库，旨在帮助开发者快速访问并应用经过验证的先进算法。它有效解决了从零开始训练大型神经网络所需的海量数据、漫长周期及高昂算力成本等痛点，让用户能直接利用现成模型进行迁移学习或推理部署。\n\n该资源库覆盖面极广，不仅包含图像分类、目标检测、语义分割等经典计算机视觉任务，还延伸至自然语言处理（如 Transformer 模型）、音频分析（语音转文字、音高估计）、激光雷达点云处理以及机器人运动规划等前沿领域。其独特的技术亮点在于与 MATLAB 生态的深度集成：自 R2024a 版本起，用户仅需调用简单的 `imagePretrainedNetwork` 函数即可一键加载如 GoogLeNet、SqueezeNet 等知名网络，无缝衔接后续的数据分析与仿真流程。\n\n无论是需要快速验证想法的算法工程师、专注于特定领域研究的高校科研人员，还是致力于将 AI 落地到自动驾驶或工业检测系统的企业开发者，都能从中获益。通过提供开箱即用的高质量模型，MATLAB Deep Learning","MATLAB Deep Learning Model Hub 是一个专为 MATLAB 用户打造的预训练深度学习模型资源库，旨在帮助开发者快速访问并应用经过验证的先进算法。它有效解决了从零开始训练大型神经网络所需的海量数据、漫长周期及高昂算力成本等痛点，让用户能直接利用现成模型进行迁移学习或推理部署。\n\n该资源库覆盖面极广，不仅包含图像分类、目标检测、语义分割等经典计算机视觉任务，还延伸至自然语言处理（如 Transformer 模型）、音频分析（语音转文字、音高估计）、激光雷达点云处理以及机器人运动规划等前沿领域。其独特的技术亮点在于与 MATLAB 生态的深度集成：自 R2024a 版本起，用户仅需调用简单的 `imagePretrainedNetwork` 函数即可一键加载如 GoogLeNet、SqueezeNet 等知名网络，无缝衔接后续的数据分析与仿真流程。\n\n无论是需要快速验证想法的算法工程师、专注于特定领域研究的高校科研人员，还是致力于将 AI 落地到自动驾驶或工业检测系统的企业开发者，都能从中获益。通过提供开箱即用的高质量模型，MATLAB Deep Learning Model Hub 大幅降低了深度学习的应用门槛，让专业人士能将更多精力聚焦于解决具体的业务难题而非重复的基础建模工作。","# MATLAB Deep Learning Model Hub\n\nDiscover pretrained models for deep learning in MATLAB.\n\n## Models \u003Ca name=\"Models\"\u002F>\n\n### Computer Vision\n* [Image Classification](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#image-classification-)\n* [Object Detection](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#object-detection-)\n* [Semantic Segmentation](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#semantic-segmentation-)\n* [Instance Segmentation](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#instance-segmentation-)\n* [Image Translation](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#image-translation-)\n* [Pose Estimation](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#pose-estimation-)\n* [3D Reconstruction](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#3d-reconstruction-)\n* [Video Classification](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#video-classification-)\n* [Text Detection & Recognition](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#text-detection-and-recognition-)\n\n### Natural Language Processing\n* [Transformers](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#transformers-text-)\n\n### Audio\n* [Audio Embeddings](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#audio-embeddings-)\n* [Sound Classification](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#application-specific-audio-models)\n* [Pitch Estimation](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#application-specific-audio-models)\n* [Speech to Text](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#speech-to-text-)\n\n### Lidar \n* [Point Cloud Classification](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#lidar-)\n* [Point Cloud Segmentation](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#lidar-)\n* [Point Cloud Object Detection](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#lidar-)\n\n### Robotics\n* [Manipulator Motion Planning](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#manipulator-motion-planning-)\n* [Path Planning with Motion Planning Networks](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#path-planning-with-motion-planning-networks-)\n\n## Image Classification \u003Ca name=\"ImageClassification\"\u002F>\n\nPretrained image classification networks have already learned to extract powerful and informative features from natural images. Use them as a starting point to learn a new task using transfer learning. \n\nInputs are RGB images, the output is the predicted label and score:\n\n![](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_9386a5205ec1.png)\n\nThese networks have been trained on more than a million images and can classify images into 1000 object categories. \n\n**Models available in MATLAB:**\n\n**Note 1: Since R2024a, please use the [imagePretrainedNetwork](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fimagepretrainednetwork.html) function instead and specify the pretrained model**.\nFor example, use the following code to access googlenet:\n```\n[net, classes] = imagePretrainedNetwork(\"googlenet\");\n```\n\n| Network |   Size (MB)  | Classes | Accuracy % | Location | \n| ------------- | ------------- | ------------- |  ------------- |  ------------- | \n| [googlenet](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fgooglenet.html)\u003Csup>1\u003Csup> | 27| 1000| 66.25 |[Doc](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fgooglenet.html) \u003Cbr \u002F>[GitHub](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Fgooglenet) |\n| [squeezenet](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fsqueezenet.html)\u003Csup>1\u003Csup>   | 5.2|  1000| 55.16 |[Doc](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fsqueezenet.html) |\n| [alexnet](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Falexnet.html)\u003Csup>1\u003Csup>  |  227|  1000|54.10|[Doc](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Falexnet.html) |\n| [resnet18](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fresnet18.html)\u003Csup>1\u003Csup>   |  44|  1000|69.49|[Doc](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fresnet18.html) \u003Cbr \u002F>[GitHub](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Fresnet-18) |\n| [resnet50](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fresnet50.html)\u003Csup>1\u003Csup>   |  96|  1000|74.46|[Doc](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fresnet50.html) \u003Cbr \u002F>[GitHub](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Fresnet-50) |\n| [resnet101](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fresnet101.html)\u003Csup>1\u003Csup>   |  167|  1000|75.96|[Doc](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fresnet101.html) \u003Cbr \u002F>[GitHub](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Fresnet-101) |\n| [mobilenetv2](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fmobilenetv2.html)\u003Csup>1\u003Csup>    | 13| 1000| 70.44|[Doc](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fmobilenetv2.html) \u003Cbr \u002F>[GitHub](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Fmobilenet-v2) |\n| [vgg16](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fvgg16.html)\u003Csup>1\u003Csup>    |  515|  1000|70.29|[Doc](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fvgg16.html) |\n| [vgg19](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fvgg19.html)\u003Csup>1\u003Csup>    |  535|  1000|70.42|[Doc](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fvgg19.html) |\n| [inceptionv3](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Finceptionv3.html)\u003Csup>1\u003Csup> | 89|  1000|77.07|[Doc](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Finceptionv3.html) |\n| [inceptionresnetv2](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Finceptionresnetv2.html)\u003Csup>1\u003Csup>  | 209|  1000|79.62|[Doc](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Finceptionresnetv2.html) |\n| [xception](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fxception.html)\u003Csup>1\u003Csup>   |  85|  1000|78.20|[Doc](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fxception.html) |\n| [darknet19](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fdarknet19.html)\u003Csup>1\u003Csup>   |  78| 1000|74.00|[Doc](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fdarknet19.html) |\n| [darknet53](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fdarknet53.html)\u003Csup>1\u003Csup>    |  155|  1000|76.46|[Doc](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fdarknet53.html) |\n| [densenet201](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fdensenet201.html)\u003Csup>1\u003Csup>    | 77| 1000| 75.85|[Doc](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fdensenet201.html) |\n| [shufflenet](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fshufflenet.html)\u003Csup>1\u003Csup>   | 5.4|  1000|63.73|[Doc](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fshufflenet.html) |\n| [nasnetmobile](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fnasnetmobile.html)\u003Csup>1\u003Csup>    | 20|  1000|73.41|[Doc](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fnasnetmobile.html) |\n| [nasnetlarge](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fnasnetlarge.html)\u003Csup>1\u003Csup>    |  332| 1000|81.83|[Doc](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fnasnetlarge.html) |\n| [efficientnetb0](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fefficientnetb0.html)\u003Csup>1\u003Csup>   | 20|  1000|74.72|[Doc](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fefficientnetb0.html) |\n| [ConvMixer](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Fconvmixer-patches-are-all-you-need)   | 7.7|  10|-|[GitHub](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Fconvmixer-patches-are-all-you-need) |\n| [Vison Transformer](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fref\u002Fvisiontransformer.html)   | Large-16 - 1100\u003Cbr \u002F> Base-16 - 331.4\u003Cbr \u002F> Small-16 - 84.7\u003Cbr \u002F> Tiny-16 - 22.2|  1000|Large-16 - 85.59\u003Cbr \u002F> Base-16 - 85.49\u003Cbr \u002F> Small-16 - 83.73\u003Cbr \u002F> Tiny-16 - 78.22|[Doc](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fref\u002Fvisiontransformer.html) |\n\n**Tips for selecting a model**\n\nPretrained networks have different characteristics that matter when choosing a network to apply to your problem. The most important characteristics are network accuracy, speed, and size. Choosing a network is generally a tradeoff between these characteristics. The following figure highlights these tradeoffs:\n\n![](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_c8acfeaa6d45.png)\nFigure. Comparing image classification model accuracy, speed and size.\n\n[Back to top](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#matlab-deep-learning-model-hub)\n\n## Object Detection \u003Ca name=\"ObjectDetection\"\u002F>\n\nObject detection is a computer vision technique used for locating instances of objects in images or videos. When humans look at images or video, we can recognize and locate objects of interest within a matter of moments. The goal of object detection is to replicate this intelligence using a computer.\n\nInputs are RGB images, the output is the predicted label, bounding box and score:\n\n![](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_c0767ac12bf5.png)\n\nThese networks have been trained to detect 80 objects classes from the COCO dataset. These models are suitable for training a custom object detector using transfer learning.\n\n| Network  | Network variants |  Size (MB) | Mean Average Precision (mAP) |Object Classes | Location |\n| ------------- | ------------- |------------ | ------------- |------------ |------------ |\n| [EfficientDet-D0](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Fpretrained-efficientdet-d0) | efficientnet |15.9 | 33.7 |80 |[GitHub](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Fpretrained-efficientdet-d0) | \n| [YOLO v9](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FPretrained-Yolov9-Network-For-Object-Detection\u002F) | yolo9t\u003Cbr \u002F>yolo9s\u003Cbr \u002F>yolo9m\u003Cbr \u002F>yolo9c\u003Cbr \u002F>yolo9e | 7.5 \u003Cbr \u002F> 25 \u003Cbr \u002F> 67.2 \u003Cbr \u002F> 85 \u003Cbr \u002F>190 | 38.3\u003Cbr \u002F> 46.8\u003Cbr \u002F> 51.4\u003Cbr \u002F>53.0 \u003Cbr \u002F>55.6|80 |[GitHub](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FPretrained-Yolov9-Network-For-Object-Detection\u002F)| \n| [YOLO v8](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FPretrained-YOLOv8-Network-For-Object-Detection) | yolo8n\u003Cbr \u002F>yolo8s\u003Cbr \u002F>yolo8m\u003Cbr \u002F>yolo8l\u003Cbr \u002F>yolo8x | 10.7 \u003Cbr \u002F> 37.2\u003Cbr \u002F>85.4 \u003Cbr \u002F>143.3\u003Cbr \u002F>222.7 | 37.3\u003Cbr \u002F>44.9\u003Cbr \u002F>50.2\u003Cbr \u002F>52.9\u003Cbr \u002F>53.9|80 |[GitHub](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FPretrained-YOLOv8-Network-For-Object-Detection)| \n| [YOLOX](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fref\u002Fyoloxobjectdetector.html) | YoloX-s\u003Cbr \u002F>YoloX-m\u003Cbr \u002F>YoloX-l | 32 \u003Cbr \u002F> 90.2\u003Cbr \u002F>192.9 | 39.8 \u003Cbr \u002F>45.9\u003Cbr \u002F>48.6|80 |[Doc](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fref\u002Fyoloxobjectdetector.html)\u003Cbr \u002F>[GitHub](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FPretrained-YOLOX-Network-For-Object-Detection)| \n| [YOLO v4](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fref\u002Fyolov4objectdetector.html) | yolov4-coco \u003Cbr \u002F> yolov4-tiny-coco| 229 \u003Cbr \u002F> 21.5 | 44.2 \u003Cbr \u002F>19.7|80 |[Doc](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fref\u002Fyolov4objectdetector.html)\u003Cbr \u002F>[GitHub](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Fpretrained-yolo-v4)| \n| [YOLO v3](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fref\u002Fyolov3objectdetector.html)| darknet53-coco \u003Cbr \u002F> tiny-yolov3-coco | 220.4 \u003Cbr \u002F> 31.5 | 34.4 \u003Cbr \u002F> 9.3 |80 |[Doc](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fref\u002Fyolov3objectdetector.html) |\n| [YOLO v2](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fref\u002Fyolov2objectdetector.html)   | darknet19-COCO \u003Cbr \u002F>tiny-yolo_v2-coco|181 \u003Cbr \u002F> 40 | 28.7 \u003Cbr \u002F> 10.5 |80 |[Doc](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fref\u002Fyolov2objectdetector.html)\u003Cbr \u002F>[GitHub](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FObject-Detection-Using-Pretrained-YOLO-v2)|\n\n**Tips for selecting a model**\n\nPretrained object detectors have different characteristics that matter when choosing a network to apply to your problem. The most important characteristics are mean average precision (mAP), speed, and size. Choosing a network is generally a tradeoff between these characteristics.\n\n**Application Specific Object Detectors**\n\nThese networks have been trained to detect specific objects for a given application.\n| Network  | Application |Size (MB) |Location |Example Output |\n| ------------- | ------------- | ------------- |------------- |------------- |\n| [Spatial-CNN](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Fpretrained-spatial-CNN)   | Lane detection | 74 |[GitHub ](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Fpretrained-spatial-CNN)|\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_5bbca0cd9368.jpg\" width=150>|\n| [RESA](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FPretrained-RESA-Network-For-Road-Boundary-Detection)   | Road Boundary detection | 95 |[GitHub ](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FPretrained-RESA-Network-For-Road-Boundary-Detection)|\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_88e39b25d92b.png\" width=150>|\n| [Single Shot Detector (SSD)](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fug\u002Fobject-detection-using-single-shot-detector.html)   | Vehicle detection | 44 |[Doc ](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fug\u002Fobject-detection-using-single-shot-detector.html)|\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_b5240c4da819.png\" width=150>|\n| [Faster R-CNN](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fug\u002Fobject-detection-using-faster-r-cnn-deep-learning.html)   | Vehicle detection | 118 |[Doc](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fug\u002Fobject-detection-using-faster-r-cnn-deep-learning.html)|\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_160ff98e9aaa.png\" width=150>|\n\n[Back to top](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#matlab-deep-learning-model-hub)\n\n## Semantic Segmentation \u003Ca name=\"SemanticSegmentation\"\u002F>\n\nSegmentation is essential for image analysis tasks. Semantic segmentation describes the process of associating each pixel of an image with a class label, (such as flower, person, road, sky, ocean, or car). \n\nInputs are RGB images, outputs are pixel classifications (semantic maps).\n\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_a4a4cb11250d.png\" class=\"center\"> \n\nThis network has been trained to detect 20 objects classes from the PASCAL VOC dataset:\n\n| Network  | Size (MB)| Mean Accuracy | Object Classes| Location | \n| ------------- | ------------- |------------- |------------- |------------- |\n| [DeepLabv3+](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Fpretrained-deeplabv3plus)   | 209 | 0.87 | 20 | [GitHub](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Fpretrained-deeplabv3plus) |\n\n Zero-shot image segmentation model:\n\n| Network  | Size (MB) | Example Location |\n| ------------- | ------------- |------------- |\n| [segmentAnythingModel](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fimages\u002Fref\u002Fsegmentanythingmodel.html)  | 358 | [Doc](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fimages\u002Fgetting-started-with-segment-anything-model.html) |\n\n\n**Application Specific Semantic Segmentation Models**\n| Network  | Application |Size (MB) |Location |Example Output |\n| ------------- | ------------- | ------------- |------------- |------------- |\n| [U-net](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fimages\u002Fdevelop-raw-camera-processing-pipeline-using-deep-learning.html)   | Raw Camera Processing |31 |[Doc](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fimages\u002Fdevelop-raw-camera-processing-pipeline-using-deep-learning.html) | \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_ee44c14e7371.png\" width=150>|\n| [3-D U-net](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fimages\u002Fsegment-3d-brain-tumor-using-deep-learning.html) | Brain Tumor Segmentation  | 56.2 | [Doc](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fimages\u002Fsegment-3d-brain-tumor-using-deep-learning.html) | \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_d1741b8847fc.gif\" width=150>|\n| [AdaptSeg (GAN)](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fug\u002Ftrain-deep-learning-semantic-segmentation-network-using-3d-simulation-data.html)  | Model tuning using 3-D simulation data | 54.4 | [Doc](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fug\u002Ftrain-deep-learning-semantic-segmentation-network-using-3d-simulation-data.html) |\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_4461c55c72da.png\" width=150>|\n\n\n[Back to top](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#matlab-deep-learning-model-hub)\n\n## Instance Segmentation \u003Ca name=\"InstanceSegmentation\"\u002F>\n\nInstance segmentation is an enhanced type of object detection that generates a segmentation map for each detected instance of an object. Instance segmentation treats individual objects as distinct entities, regardless of the class of the objects. In contrast, semantic segmentation considers all objects of the same class as belonging to a single entity.\n\nInputs are RGB images, outputs are pixel classifications (semantic maps), bounding boxes and classification labels.\n\n![](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_e7cd16135607.png)\n\n| Network | Object Classes | Location |\n| ------------- | ------------- |------------- |\n| [Mask R-CNN](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fref\u002Fmaskrcnn.html) | 80 | [Doc](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fref\u002Fmaskrcnn.html) \u003Cbr \u002F>  [Github](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Fmask-rcnn)|\n\n[Back to top](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#matlab-deep-learning-model-hub)\n\n## Image Translation \u003Ca name=\"ImageTranslation\"\u002F>\n\nImage translation is the task of transferring styles and characteristics from one image domain to another. This technique can be extended to other image-to-image learning operations, such as image enhancement, image colorization, defect generation, and medical image analysis.\n\nInputs are images, outputs are translated RGB images. This example workflow shows how a semantic segmentation map input translates to a synthetic image via a pretrained model (Pix2PixHD):\n\n![](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_47c23d3a55ea.png)\n\n| Network  | Application |Size (MB) |Location |Example Output |\n| ------------- | ------------- | ------------- |------------- |------------- |\n| [Pix2PixHD(CGAN)](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fug\u002Fgenerate-image-from-segmentation-map-using-deep-learning.html) | Synthetic Image Translation | 648 | [Doc](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fug\u002Fgenerate-image-from-segmentation-map-using-deep-learning.html) |\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_6f1b9d0a344a.png\" width=150> |\n| [UNIT (GAN)](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fimages\u002Funsupervised-day-to-dusk-image-translation-using-unit.html) | Day-to-Dusk Dusk-to-Day Image Translation | 72.5 | [Doc](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fimages\u002Funsupervised-day-to-dusk-image-translation-using-unit.html) |\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_fd6cbfcbd31b.png\" width=150>|\n| [UNIT (GAN)](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fimages\u002Funsupervised-medical-image-denoising-using-unit.html) | Medical Image Denoising | 72.4 | [Doc](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fimages\u002Funsupervised-medical-image-denoising-using-unit.html) |\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_3a3cc0bcbba0.png\" width=150>|\n| [CycleGAN](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fimages\u002Funsupervised-medical-image-denoising-using-cyclegan.html) | Medical Image Denoising | 75.3 | [Doc](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fimages\u002Funsupervised-medical-image-denoising-using-cyclegan.html) |\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_8d11d1b7fee0.png\" width=150>|\n| [VDSR](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fimages\u002Fsingle-image-super-resolution-using-deep-learning.html) | Super Resolution (estimate a high-resolution image from a low-resolution image) | 2.4 | [Doc](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fimages\u002Fsingle-image-super-resolution-using-deep-learning.html) |\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_0334b72b3fb6.png\" width=150>|\n\n[Back to top](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#matlab-deep-learning-model-hub)\n\n## Pose Estimation \u003Ca name=\"PoseEstimation\"\u002F>\n\nPose estimation is a computer vision technique for localizing the position and orientation of an object using a fixed set of keypoints. \n\nAll inputs are RGB images, outputs are heatmaps and part affinity fields (PAFs) which via post processing perform pose estimation.\n\n![](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_dbc77f15b87a.png)\n\n| Network  | Backbone Networks | Size (MB)| Location|\n| ------------- | -------------| ------------- |------------- |\n| [OpenPose](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fug\u002Festimate-body-pose-using-deep-learning.html) | vgg19  | 14 | [Doc](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fug\u002Festimate-body-pose-using-deep-learning.html) |\n| [HR Net](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fref\u002Fhrnetobjectkeypointdetector.html) | human-full-body-w32\u003Cbr \u002F>human-full-body-w48  | 106.9\u003Cbr \u002F>237.7 | [Doc](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fref\u002Fhrnetobjectkeypointdetector.html) |\n\n\n[Back to top](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#matlab-deep-learning-model-hub)\n## 3D Reconstruction \u003Ca name=\"3DReconstruction\"\u002F>\n\n3D reconstruction is the process of capturing the shape and appearance of real objects.\n\n| Network  | Size (MB)| Location|Example Output|\n| ------------- | ------------- |------------- |------------- |\n| [NeRF](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Fnerf)   | 3.78 |[GitHub](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Fnerf)|![NeRF](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_bed99f087597.jpg) |\n\n[Back to top](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#matlab-deep-learning-model-hub)\n\n## Video Classification \u003Ca name=\"VideoClassification\"\u002F>\n\nVideo classification is a computer vision technique for classifying the action or content in a sequence of video frames. \n\nAll inputs are Videos only or Video with Optical Flow data, outputs are gesture classifications and scores.\n\n![](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_8213a60b95e2.png)\n\n| Network  | Inputs | Size(MB) | Classifications (Human Actions)| Description | Location|\n| ------------- | ------------- |------------- |------------- |------------- |------------- |\n| [SlowFast](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fref\u002Fslowfastvideoclassifier.html) | Video | 124 |400 |Faster convergence than Inflated-3D |[Doc](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fref\u002Fslowfastvideoclassifier.html)\n| [R(2+1)D](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fref\u002Fr2plus1dvideoclassifier.html) | Video | 112 |400 |Faster convergence than Inflated-3D|[Doc](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fref\u002Fr2plus1dvideoclassifier.html)\n| [Inflated-3D](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fref\u002Finflated3dvideoclassifier.html) | Video & Optical Flow data | 91 | 400 |Accuracy of the classifier improves when combining optical flow and RGB data.| [Doc](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fref\u002Finflated3dvideoclassifier.html)\n\n[Back to top](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#matlab-deep-learning-model-hub)\n\n## Text Detection and Recognition \u003Ca name=\"textdetection\"\u002F>\n\nText detection is a computer vision technique used for locating instances of text within in images.\n\nInputs are RGB images, outputs are bounding boxes that identify regions of text.\n\n![](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_5d0127dedf7d.png)\n\n| Network  | Application | Size (MB) | Location |\n| ------------- | ------------- |------------- |------------- |\n| [CRAFT](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fref\u002Fdetecttextcraft.html)   | Trained to detect English, Korean, Italian, French, Arabic, German and Bangla (Indian).| 3.8 |[Doc](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fref\u002Fdetecttextcraft.html) \u003Cbr \u002F> [GitHub](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FText-Detection-using-Deep-Learning) |\n\n**Application Specific Text Detectors**\n\n| Network  | Application | Size (MB) | Location |Example Output |\n| ------------- | ------------- |------------- |------------- |------------- |\n| [Seven Segment Digit Recognition](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FSeven-Segment-Digit-Recognition)   |Seven segment digit recognition using deep learning and OCR. This is helpful in industrial automation applications where digital displays are often surrounded with complex background. | 3.8 |[Doc](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fref\u002Focr.html) \u003Cbr \u002F> [GitHub](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FSeven-Segment-Digit-Recognition)  |![](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_2e64190ead1d.png)|\n\n[Back to top](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#matlab-deep-learning-model-hub)\n\n## Transformers (Text) \u003Ca name=\"transformers\"\u002F>\n\nTransformer pretained models have already learned to extract powerful and informative features features from text. Use them as a starting point to learn a new task using transfer learning. \n\nInputs are sequences of text, outputs are text feature embeddings.\n\n![](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_2502418d7649.png)\n\n\n| Network  | Applications | Size (MB) | Location |\n| ------------- | ------------- |------------ |------------ |\n| [BERT](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Ftextanalytics\u002Fref\u002Fbert.html)   | Feature Extraction (Sentence and Word embedding), Text Classification, Token Classification, Masked Language Modeling, Question Answering |390 |[GitHub](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Ftransformer-models#bert-and-finbert) \u003Cbr \u002F> [Doc](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Ftextanalytics\u002Fref\u002Fbert.html) | \n| [all-MiniLM-L6-v2](https:\u002F\u002Fwww.mathworks.com\u002Fmatlabcentral\u002Ffileexchange\u002F156399-text-analytics-toolbox-model-for-all-minilm-l6-v2-network)   | Document Embedding, Clustering, Information Retrieval  |80 |[Doc](https:\u002F\u002Fwww.mathworks.com\u002Fmatlabcentral\u002Ffileexchange\u002F156399-text-analytics-toolbox-model-for-all-minilm-l6-v2-network)  | \n| [all-MiniLM-L12-v2](https:\u002F\u002Fwww.mathworks.com\u002Fmatlabcentral\u002Ffileexchange\u002F156394-text-analytics-toolbox-model-for-all-minilm-l12-v2-network)   | Document Embedding, Clustering, Information Retrieval  |120 |[Doc](https:\u002F\u002Fwww.mathworks.com\u002Fmatlabcentral\u002Ffileexchange\u002F156394-text-analytics-toolbox-model-for-all-minilm-l12-v2-network)  | \n\n**Application Specific Transformers**\n\n| Network  | Application | Size (MB) | Location | Output Example |\n| ------------- | ------------- | ------------- |------------- |------------- |\n| [FinBERT](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Ftransformer-models#bert-and-finbert)   | The FinBERT model is a BERT model for financial sentiment analysis | 388 |[GitHub](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Ftransformer-models#bert-and-finbert) |![](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_6b916896ae8a.png)|\n| [GPT-2](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Ftransformer-models#gpt-2) | The GPT-2 model is a decoder model used for text summarization.| 1.2GB |[GitHub](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Ftransformer-models#gpt-2) |![](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_0694504b0d56.png)|\n\n[Back to top](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#matlab-deep-learning-model-hub)\n\n## Audio Embeddings \u003Ca name=\"AudioEmbeddings\"\u002F>\n\nAudio embedding pretrained models have already learned to extract powerful and informative features from audio signals. Use them as a starting point to learn a new task using transfer learning. \n\nInputs are audio signals, outputs are audio feature embeddings.\n\n![](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_32113253007f.png)\n\n**Note 2: Since R2024a, please use the [audioPretrainedNetwork](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Faudio\u002Fref\u002Faudiopretrainednetwork.html) function instead and specify the pretrained model**.\nFor example, use the following code to access VGGish:\n```\nnet = audioPretrainedNetwork(\"vggish\");\n```\n\n| Network  | Application | Size (MB) | Location |\n| ------------- | ------------- | ------------- |------------- |\n| [VGGish](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Faudio\u002Fref\u002Fvggish.html)\u003Csup>2\u003Csup>   | Feature Embeddings | 257 |[Doc](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Faudio\u002Fref\u002Fvggish.html) |\n| [OpenL3](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Faudio\u002Fref\u002Fopenl3.html)\u003Csup>2\u003Csup>   | Feature Embeddings | 200 |[Doc](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Faudio\u002Fref\u002Fopenl3.html) |\n\n## Application Specific Audio Models\u003Ca name=\"Application Specific Audio Models\"\u002F>\n\n| Network  | Application | Size (MB) | Output Classes | Location | Output Example |\n| ------------- | ------------- | ------------- |------------- |------------- |------------- |\n| \u003Ca name=\"SoundClassification\"\u002F>[vadnet](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Faudio\u002Fref\u002Fvadnet.html)\u003Csup>2\u003Csup>   | Voice Activity Detection (regression) | 0.427 | - |[Doc](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Faudio\u002Fref\u002Fvadnet.html) |\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_4e71efcc5aa3.png\" width=150>|\n| \u003Ca name=\"SoundClassification\"\u002F>[YAMNet](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Faudio\u002Fref\u002Fyamnet.html)\u003Csup>2\u003Csup>   | Sound Classification | 13.5 | 521 |[Doc](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Faudio\u002Fref\u002Fyamnet.html) |\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_5fb224370fce.png\" width=150>|\n| \u003Ca name=\"PitchEstimation\"\u002F>[CREPE](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Faudio\u002Fref\u002Fcrepe.html)\u003Csup>2\u003Csup>   | Pitch Estimation (regression) | 132| - |[Doc](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Faudio\u002Fref\u002Fcrepe.html) |\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_a90b8645ea33.png\" width=150>|\n\n## Speech to Text \u003Ca name=\"Speech2Text\"\u002F>\n\nSpeech-to-text models provide a fast, efficient method to convert spoken language into written text, enhancing accessibility for individuals with disabilities, enabling downstream tasks like text summarization and sentiment analysis, and streamlining documentation processes. As a key element of human-machine interfaces, including personal assistants, it allows for natural and intuitive interactions, enabling machines to understand and execute spoken commands, improving usability and broadening inclusivity across various applications.\n\nInputs are audio signals, outputs is text.\n\n![](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_34a5ee07f3d2.png)\n\n| Network  | Application | Size (MB) | Word Error Rate (WER) | Location | \n| ------------- | ------------- | ------------- |------------- |------------- |\n| [wav2vec](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Fwav2vec-2.0)   | Speech to Text | 236| 3.2 |[GitHub](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Fwav2vec-2.0) |\n| [deepspeech](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Fdeepspeech)   | Speech to Text | 167| 5.97 |[GitHub](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Fdeepspeech) |\n\n[Back to top](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#matlab-deep-learning-model-hub)\n\n## Lidar \u003Ca name=\"PointCloud\"\u002F>\n\nPoint cloud data is acquired by a variety of sensors, such as lidar, radar, and depth cameras. Training robust classifiers with point cloud data is challenging because of the sparsity of data per object, object occlusions, and sensor noise. Deep learning techniques have been shown to address many of these challenges by learning robust feature representations directly from point cloud data. \n\nInputs are Lidar Point Clouds converted to five-channels, outputs are segmentation, classification or object detection results overlayed on point clouds.\n\n![](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_bbd74a3633b7.png)\n\n| Network  | Application | Size (MB) | Object Classes | Location | \n| ------------- | ------------- | ------------- |------------- |------------- |\n| [PointNet](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fug\u002Fpoint-cloud-classification-using-pointnet-deep-learning.html)  | Classification | 5| 14 |[Doc](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fug\u002Fpoint-cloud-classification-using-pointnet-deep-learning.html)|\n| \u003Ca name=\"PointCloudSeg\"\u002F>[PointNet++](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Flidar\u002Fug\u002Faerial-lidar-segmentation-using-pointnet-network.html)  | Segmentation | 3| 8 |[Doc](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Flidar\u002Fug\u002Faerial-lidar-segmentation-using-pointnet-network.html)|\n| [PointSeg](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fug\u002Flidar-semantic-segmentation-using-pointseg.html)   | Segmentation | 14| 3 |[Doc](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fug\u002Fpoint-cloud-classification-using-pointnet-deep-learning.html)|\n| [SqueezeSegV2](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fug\u002Flidar-semantic-segmentation-using-squeezesegv2.html)   | Segmentation |5| 12 |[Doc](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fug\u002Flidar-semantic-segmentation-using-squeezesegv2.html) |\n| [SalsaNext](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Fpretrained-salsanext)   | Segmentation |20.9 | 13 |[GitHub](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Fpretrained-salsanext)|\n| \u003Ca name=\"PointCloudObj\"\u002F>[PointPillars](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Flidar\u002Fug\u002Fobject-detection-using-pointpillars-network.html)   | Object Detection | 8| 3 |[Doc](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Flidar\u002Fug\u002Fobject-detection-using-pointpillars-network.html)|\n| [Complex YOLO v4](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FLidar-object-detection-using-complex-yolov4)   | Object Detection | 233 (complex-yolov4) \u003Cbr \u002F> 21 (tiny-complex-yolov4) | 3 |[GitHub](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FLidar-object-detection-using-complex-yolov4)|\n\n[Back to top](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#matlab-deep-learning-model-hub)\n\n## Manipulator Motion Planning \u003Ca name=\"ManipMotionPlanning\"\u002F>\n\nManipulator motion planning is a technique used to plan a trajectory for a robotic arm from a start position to a goal position in an obstacle environment.\n  \nPretrained deep learning models have learned to plan such trajectories  for repetitive tasks such as picking and placing of objects, leading to speed ups over traditional algorithms.\n\nInputs are start configuration, goal configuration and obstacle environment encoding for the robot, outputs are intermediate trajectory guesses.\n\n![](Images\u002Fmotion_planning_workflow.svg)\n\n| Network  | Application | Size (MB)| Location|\n| --- | --- | --- | --- |\n| [Deep-Learning-Based CHOMP (DLCHOMP)](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Freleases\u002FR2024a\u002Frobotics\u002Fref\u002Fdlchomp.html)   | Trajectory Prediction | 25 | [Doc](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Freleases\u002FR2024a\u002Frobotics\u002Fref\u002Fdlchomp.html)\u003Cbr \u002F>[GitHub](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Fpretrained-dlchomp) |\n\n\n\n[Back to top](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#matlab-deep-learning-model-hub)\n\n\n## Path Planning with Motion Planning Networks \u003Ca name=\"PathPlanningMPNet\"\u002F>\n\nMotion Planning Networks (MPNet) is a deep-learning-based approach for finding optimal paths between a start point and goal point in motion planning problems. MPNet is a deep neural network that can be trained on multiple environments to learn optimal paths between various states in the environments. The MPNet uses this prior knowledge to,\n- Generate informed samples between two states in an unknown test environment. These samples can be used with sampling-based motion planners such as optimal rapidly-exploring random trees (RRT*) for path planning.\n- Compute collision-free path between two states in an unknown test environment. MPNet based path planner is more efficient than the classical path planners such as the RRT*.\n\nTo know more please visit [Get Started with Motion Planning Networks](https:\u002F\u002Fin.mathworks.com\u002Fhelp\u002Fnav\u002Fug\u002Fget-started-with-motion-planning-networks.html)\n\n![](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_696259714deb.png)\n\n| Network                                                                                                     | Application   | Size (MB) | Location                                                                                                |\n| ----------------------------------------------------------------------------------------------------------- | ------------- | --------- | ------------------------------------------------------------------------------------------------------- |\n| [mazeMapTrainedMPNET](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fnav\u002Fug\u002Fget-started-with-motion-planning-networks.html) | Path Planning | 0.23      | [Doc](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fnav\u002Fug\u002Ftrain-deep-learning-based-sampler-for-motion-planning.html) |\n\n[Back to top](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#matlab-deep-learning-model-hub)\n\n## Model requests\nIf you'd like to request MATLAB support for additional pretrained models, [**please create an issue from this repo**](https:\u002F\u002Fdocs.github.com\u002Fen\u002Fissues\u002Ftracking-your-work-with-issues\u002Fcreating-an-issue). \n\nAlternatively send the request through to:\n\n[Jianghao Wang](https:\u002F\u002Fwww.linkedin.com\u002Fin\u002Fjianghao-wang-896aa1a4\u002F) \u003Cbr \u002F> \nDeep Learning Product Manager \u003Cbr \u002F> \njianghaw@mathworks.com\n\nCopyright 2024, The MathWorks, Inc.\n\n","# MATLAB 深度学习模型库\n\n在 MATLAB 中发现用于深度学习的预训练模型。\n\n## 模型 \u003Ca name=\"Models\"\u002F>\n\n### 计算机视觉\n* [图像分类](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#image-classification-)\n* [目标检测](https0:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#object-detection-)\n* [语义分割](https0:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#semantic-segmentation-)\n* [实例分割](https0:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#instance-segmentation-)\n* [图像风格迁移](https0:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#image-translation-)\n* [人体姿态估计](https0:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#pose-estimation-)\n* [三维重建](https0:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#3d-reconstruction-)\n* [视频分类](https0:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#video-classification-)\n* [文本检测与识别](https0:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#text-detection-and-recognition-)\n\n### 自然语言处理\n* [Transformer 模型](https0:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#transformers-text-)\n\n### 音频\n* [音频嵌入](https0:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#audio-embeddings-)\n* [声音分类](https0:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#application-specific-audio-models)\n* [音高估计](https0:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#application-specific-audio-models)\n* [语音转文本](https0:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#speech-to-text-)\n\n### 激光雷达\n* [点云分类](https0:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#lidar-)\n* [点云分割](https0:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#lidar-)\n* [点云目标检测](https0:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#lidar-)\n\n### 机器人技术\n* [机械臂运动规划](https0:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#manipulator-motion-planning-)\n* [基于运动规划网络的路径规划](https0:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#path-planning-with-motion-planning-networks-)\n\n## 图像分类 \u003Ca name=\"ImageClassification\"\u002F>\n\n预训练的图像分类网络已经学会了从自然图像中提取强大且富含信息的特征。可以将它们作为起点，利用迁移学习来学习新的任务。\n\n输入是 RGB 图像，输出是预测的标签和得分：\n\n![](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_9386a5205ec1.png)\n\n这些网络已在超过一百万张图像上进行过训练，能够将图像分类到 1000 个对象类别中。\n\n**MATLAB 中可用的模型：**\n\n**注 1：自 R2024a 起，请改用 [imagePretrainedNetwork](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fimagepretrainednetwork.html) 函数，并指定预训练模型**。\n例如，使用以下代码访问 googlenet：\n```\n[net, classes] = imagePretrainedNetwork(\"googlenet\");\n```\n\n| 网络 | 大小 (MB) | 类别数 | 准确率 (%) | 获取方式 |\n| ------------- | ------------- | ------------- |  ------------- |  ------------- | \n| [googlenet](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fgooglenet.html)\u003Csup>1\u003Csup> | 27| 1000| 66.25 |[文档](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fgooglenet.html) \u003Cbr \u002F>[GitHub](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Fgooglenet) |\n| [squeezenet](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fsqueezenet.html)\u003Csup>1\u003Csup>   | 5.2|  1000| 55.16 |[文档](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fsqueezenet.html) |\n| [alexnet](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Falexnet.html)\u003Csup>1\u003Csup>  |  227|  1000|54.10|[文档](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Falexnet.html) |\n| [resnet18](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fresnet18.html)\u003Csup>1\u003Csup>   |  44|  1000|69.49|[文档](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fresnet18.html) \u003Cbr \u002F>[GitHub](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Fresnet-18) |\n| [resnet50](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fresnet50.html)\u003Csup>1\u003Csup>   |  96|  1000|74.46|[文档](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fresnet50.html) \u003Cbr \u002F>[GitHub](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Fresnet-50) |\n| [resnet101](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fresnet101.html)\u003Csup>1\u003Csup>   |  167|  1000|75.96|[文档](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fresnet101.html) \u003Cbr \u002F>[GitHub](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Fresnet-101) |\n| [mobilenetv2](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fmobilenetv2.html)\u003Csup>1\u003Csup>    | 13| 1000| 70.44|[文档](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fmobilenetv2.html) \u003Cbr \u002F>[GitHub](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Fmobilenet-v2) |\n| [vgg16](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fvgg16.html)\u003Csup>1\u003Csup>    |  515|  1000|70.29|[文档](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fvgg16.html) |\n| [vgg19](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fvgg19.html)\u003Csup>1\u003Csup>    |  535|  1000|70.42|[文档](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fvgg19.html) |\n| [inceptionv3](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Finceptionv3.html)\u003Csup>1\u003Csup> | 89|  1000|77.07|[文档](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Finceptionv3.html) |\n| [inceptionresnetv2](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Finceptionresnetv2.html)\u003Csup>1\u003Csup>  | 209|  1000|79.62|[文档](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Finceptionresnetv2.html) |\n| [xception](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fxception.html)\u003Csup>1\u003Csup>   |  85|  1000|78.20|[文档](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fxception.html) |\n| [darknet19](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fdarknet19.html)\u003Csup>1\u003Csup>   |  78| 1000|74.00|[文档](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fdarknet19.html) |\n| [darknet53](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fdarknet53.html)\u003Csup>1\u003Csup>    |  155|  1000|76.46|[文档](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fdarknet53.html) |\n| [densenet201](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fdensenet201.html)\u003Csup>1\u003Csup>    | 77| 1000| 75.85|[文档](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fdensenet201.html) |\n| [shufflenet](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fshufflenet.html)\u003Csup>1\u003Csup>   | 5.4|  1000|63.73|[文档](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fshufflenet.html) |\n| [nasnetmobile](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fnasnetmobile.html)\u003Csup>1\u003Csup>    | 20|  1000|73.41|[文档](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fnasnetmobile.html) |\n| [nasnetlarge](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fnasnetlarge.html)\u003Csup>1\u003Csup>    |  332| 1000|81.83|[文档](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fnasnetlarge.html) |\n| [efficientnetb0](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fefficientnetb0.html)\u003Csup>1\u003Csup>   | 20|  1000|74.72|[文档](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fefficientnetb0.html) |\n| [ConvMixer](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Fconvmixer-patches-are-all-you-need)   | 7.7|  10|-|[GitHub](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Fconvmixer-patches-are-all-you-need) |\n| [Vision Transformer](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fref\u002Fvisiontransformer.html)   | Large-16 - 1100\u003Cbr \u002F> Base-16 - 331.4\u003Cbr \u002F> Small-16 - 84.7\u003Cbr \u002F> Tiny-16 - 22.2|  1000|Large-16 - 85.59\u003Cbr \u002F> Base-16 - 85.49\u003Cbr \u002F> Small-16 - 83.73\u003Cbr \u002F> Tiny-16 - 78.22|[文档](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fref\u002Fvisiontransformer.html) |\n\n**选择模型的建议**\n\n预训练网络具有不同的特性，在为您的问题选择网络时这些特性非常重要。最重要的特性包括网络的准确率、速度和大小。通常，选择网络需要在这几者之间进行权衡。下图展示了这些权衡关系：\n\n![](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_c8acfeaa6d45.png)\n图. 比较图像分类模型的准确率、速度和大小。\n\n[返回顶部](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#matlab-deep-learning-model-hub)\n\n## 目标检测 \u003Ca name=\"ObjectDetection\"\u002F>\n\n目标检测是一种计算机视觉技术，用于在图像或视频中定位物体实例。当人类观看图像或视频时，我们能够在瞬间识别并定位感兴趣的物体。目标检测的目标是利用计算机来复制这种智能。\n\n输入为RGB图像，输出为预测的标签、边界框和置信度分数：\n\n![](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_c0767ac12bf5.png)\n\n这些网络已经过训练，能够检测COCO数据集中的80个物体类别。这些模型适合通过迁移学习来训练自定义的目标检测器。\n\n| 网络  | 网络变体 | 大小 (MB) | 平均精度均值 (mAP) | 物体类别 | 获取方式 |\n| ------------- | ------------- |------------ | ------------- |------------ |------------ |\n| [EfficientDet-D0](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Fpretrained-efficientdet-d0) | efficientnet |15.9 | 33.7 |80 |[GitHub](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Fpretrained-efficientdet-d0) | \n| [YOLO v9](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FPretrained-Yolov9-Network-For-Object-Detection\u002F) | yolo9t\u003Cbr \u002F>yolo9s\u003Cbr \u002F>yolo9m\u003Cbr \u002F>yolo9c\u003Cbr \u002F>yolo9e | 7.5 \u003Cbr \u002F> 25 \u003Cbr \u002F> 67.2 \u003Cbr \u002F> 85 \u003Cbr \u002F>190 | 38.3\u003Cbr \u002F> 46.8\u003Cbr \u002F> 51.4\u003Cbr \u002F>53.0 \u003Cbr \u002F>55.6|80 |[GitHub](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FPretrained-Yolov9-Network-For-Object-Detection\u002F)| \n| [YOLO v8](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FPretrained-YOLOv8-Network-For-Object-Detection) | yolo8n\u003Cbr \u002F>yolo8s\u003Cbr \u002F>yolo8m\u003Cbr \u002F>yolo8l\u003Cbr \u002F>yolo8x | 10.7 \u003Cbr \u002F> 37.2\u003Cbr \u002F>85.4 \u003Cbr \u002F>143.3\u003Cbr \u002F>222.7 | 37.3\u003Cbr \u002F>44.9\u003Cbr \u002F>50.2\u003Cbr \u002F>52.9\u003Cbr \u002F>53.9|80 |[GitHub](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FPretrained-YOLOv8-Network-For-Object-Detection)| \n| [YOLOX](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fref\u002Fyoloxobjectdetector.html) | YoloX-s\u003Cbr \u002F>YoloX-m\u003Cbr \u002F>YoloX-l | 32 \u003Cbr \u002F> 90.2\u003Cbr \u002F>192.9 | 39.8 \u003Cbr \u002F>45.9\u003Cbr \u002F>48.6|80 |[文档](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fref\u002Fyoloxobjectdetector.html)\u003Cbr \u002F>[GitHub](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FPretrained-YOLOX-Network-For-Object-Detection)| \n| [YOLO v4](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fref\u002Fyolov4objectdetector.html) | yolov4-coco \u003Cbr \u002F> yolov4-tiny-coco| 229 \u003Cbr \u002F> 21.5 | 44.2 \u003Cbr \u002F>19.7|80 |[文档](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fref\u002Fyolov4objectdetector.html)\u003Cbr \u002F>[GitHub](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Fpretrained-yolo-v4)| \n| [YOLO v3](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fref\u002Fyolov3objectdetector.html)| darknet53-coco \u003Cbr \u002F> tiny-yolov3-coco | 220.4 \u003Cbr \u002F> 31.5 | 34.4 \u003Cbr \u002F> 9.3 |80 |[文档](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fref\u002Fyolov3objectdetector.html) |\n| [YOLO v2](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fref\u002Fyolov2objectdetector.html)   | darknet19-COCO \u003Cbr \u002F>tiny-yolo_v2-coco|181 \u003Cbr \u002F> 40 | 28.7 \u003Cbr \u002F> 10.5 |80 |[文档](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fref\u002Fyolov2objectdetector.html)\u003Cbr \u002F>[GitHub](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FObject-Detection-Using-Pretrained-YOLO-v2)|\n\n**选择模型的建议**\n\n预训练的目标检测器具有不同的特性，在为您的问题选择网络时需要考虑这些特性。最重要的特性包括平均精度均值（mAP）、速度和模型大小。通常，选择网络是在这些特性之间进行权衡。\n\n**特定应用的目标检测器**\n\n这些网络已针对特定应用进行了训练，以检测特定物体。\n| 网络  | 应用 |大小 (MB) |获取方式 |示例输出 |\n| ------------- | ------------- | ------------- |------------- |------------- |\n| [Spatial-CNN](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Fpretrained-spatial-CNN)   | 车道检测 | 74 |[GitHub ](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Fpretrained-spatial-CNN)|\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_5bbca0cd9368.jpg\" width=150>|\n| [RESA](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FPretrained-RESA-Network-For-Road-Boundary-Detection)   | 道路边界检测 | 95 |[GitHub ](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FPretrained-RESA-Network-For-Road-Boundary-Detection)|\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_88e39b25d92b.png\" width=150>|\n| [单次检测器 (SSD)](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fug\u002Fobject-detection-using-single-shot-detector.html)   | 车辆检测 | 44 |[文档 ](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fug\u002Fobject-detection-using-single-shot-detector.html)|\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_b5240c4da819.png\" width=150>|\n| [Faster R-CNN](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fug\u002Fobject-detection-using-faster-r-cnn-deep-learning.html)   | 车辆检测 | 118 |[文档](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fug\u002Fobject-detection-using-faster-r-cnn-deep-learning.html)|\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_160ff98e9aaa.png\" width=150>|\n\n[返回顶部](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#matlab-deep-learning-model-hub)\n\n## 语义分割 \u003Ca name=\"SemanticSegmentation\"\u002F>\n\n分割在图像分析任务中至关重要。语义分割是指为图像中的每个像素分配一个类别标签的过程，例如花、人、道路、天空、海洋或汽车。\n\n输入是RGB图像，输出是像素分类（语义图）。\n\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_a4a4cb11250d.png\" class=\"center\"> \n\n该网络已训练用于检测PASCAL VOC数据集中的20个物体类别：\n\n| 网络  | 大小 (MB)| 平均准确率 | 物体类别| 位置 | \n| ------------- | ------------- |------------- |------------- |------------- |\n| [DeepLabv3+]（https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Fpretrained-deeplabv3plus）   | 209 | 0.87 | 20 | [GitHub](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Fpretrained-deeplabv3plus) |\n\n零样本图像分割模型：\n\n| 网络  | 大小 (MB) | 示例位置 |\n| ------------- | ------------- |------------- |\n| [segmentAnythingModel]（https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fimages\u002Fref\u002Fsegmentanythingmodel.html）  | 358 | [文档](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fimages\u002Fgetting-started-with-segment-anything-model.html) |\n\n\n**特定应用的语义分割模型**\n| 网络  | 应用 |大小 (MB) |位置 |示例输出 |\n| ------------- | ------------- | ------------- |------------- |------------- |\n| [U-net]（https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fimages\u002Fdevelop-raw-camera-processing-pipeline-using-deep-learning.html）   | 原始图像处理 |31 |[文档](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fimages\u002Fdevelop-raw-camera-processing-pipeline-using-deep-learning.html) | \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_ee44c14e7371.png\" width=150>|\n| [3-D U-net]（https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fimages\u002Fsegment-3d-brain-tumor-using-deep-learning.html） | 脑肿瘤分割  | 56.2 | [文档](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fimages\u002Fsegment-3d-brain-tumor-using-deep-learning.html) | \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_d1741b8847fc.gif\" width=150>|\n| [AdaptSeg (GAN)]（https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fug\u002Ftrain-deep-learning-semantic-segmentation-network-using-3d-simulation-data.html）  | 使用3D仿真数据进行模型调优 | 54.4 | [文档](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fug\u002Ftrain-deep-learning-semantic-segmentation-network-using-3d-simulation-data.html) |\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_4461c55c72da.png\" width=150>|\n\n\n[返回顶部](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#matlab-deep-learning-model-hub)\n\n## 实例分割 \u003Ca name=\"InstanceSegmentation\"\u002F>\n\n实例分割是一种增强型目标检测技术，它为每个检测到的对象实例生成一个分割掩码。实例分割将各个对象视为独立的实体，而不考虑它们所属的类别。相比之下，语义分割则将同一类别的所有对象视为一个整体。\n\n输入是RGB图像，输出是像素分类（语义图）、边界框和类别标签。\n\n![](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_e7cd16135607.png)\n\n| 网络 | 物体类别 | 位置 |\n| ------------- | ------------- |------------- |\n| [Mask R-CNN]（https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fref\u002Fmaskrcnn.html） | 80 | [文档](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fref\u002Fmaskrcnn.html) \u003Cbr \u002F>  [GitHub](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Fmask-rcnn)|\n\n[返回顶部](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#matlab-deep-learning-model-hub)\n\n## 图像翻译 \u003Ca name=\"ImageTranslation\"\u002F>\n\n图像翻译是指将一种图像域的风格和特征迁移到另一种图像域的任务。这一技术还可以扩展到其他图像到图像的学习任务，如图像增强、图像着色、缺陷生成以及医学图像分析等。\n\n输入是图像，输出是经过翻译后的RGB图像。此示例工作流展示了如何通过预训练模型（Pix2PixHD）将语义分割图输入转换为合成图像：\n\n![](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_47c23d3a55ea.png)\n\n| 网络  | 应用 |大小 (MB) |位置 |示例输出 |\n| ------------- | ------------- | ------------- |------------- |------------- |\n| [Pix2PixHD(CGAN)]（https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fug\u002Fgenerate-image-from-segmentation-map-using-deep-learning.html） | 合成图像翻译 | 648 | [文档](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fug\u002Fgenerate-image-from-segmentation-map-using-deep-learning.html) |\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_6f1b9d0a344a.png\" width=150> |\n| [UNIT (GAN)]（https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fimages\u002Funsupervised-day-to-dusk-image-translation-using-unit.html） | 白天到黄昏、黄昏到白天的图像翻译 | 72.5 | [文档](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fimages\u002Funsupervised-day-to-dusk-image-translation-using-unit.html) |\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_fd6cbfcbd31b.png\" width=150>|\n| [UNIT (GAN)]（https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fimages\u002Funsupervised-medical-image-denoising-using-unit.html） | 医学图像去噪 | 72.4 | [文档](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fimages\u002Funsupervised-medical-image-denoising-using-unit.html) |\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_3a3cc0bcbba0.png\" width=150>|\n| [CycleGAN]（https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fimages\u002Funsupervised-medical-image-denoising-using-cyclegan.html） | 医学图像去噪 | 75.3 | [文档](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fimages\u002Funsupervised-medical-image-denoising-using-cyclegan.html) |\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_8d11d1b7fee0.png\" width=150>|\n| [VDSR]（https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fimages\u002Fsingle-image-super-resolution-using-deep-learning.html） | 超分辨率（从低分辨率图像估计高分辨率图像） | 2.4 | [文档](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fimages\u002Fsingle-image-super-resolution-using-deep-learning.html) |\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_0334b72b3fb6.png\" width=150>|\n\n[返回顶部](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#matlab-deep-learning-model-hub)\n\n## 姿态估计 \u003Ca name=\"PoseEstimation\"\u002F>\n\n姿态估计是一种计算机视觉技术，它使用一组固定的关节点来定位物体的位置和方向。\n\n所有输入都是RGB图像，输出是热图和部分亲和场（PAFs），这些输出经过后处理即可完成姿态估计。\n\n![](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_dbc77f15b87a.png)\n\n| 网络  | 主干网络 | 大小 (MB)| 位置|\n| ------------- | -------------| ------------- |------------- |\n| [OpenPose]（https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fug\u002Festimate-body-pose-using-deep-learning.html） | vgg19  | 14 | [文档](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fug\u002Festimate-body-pose-using-deep-learning.html) |\n| [HR Net]（https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fref\u002Fhrnetobjectkeypointdetector.html） | human-full-body-w32\u003Cbr \u002F>human-full-body-w48  | 106.9\u003Cbr \u002F>237.7 | [文档](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fref\u002Fhrnetobjectkeypointdetector.html) |\n\n\n[返回顶部](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#matlab-deep-learning-model-hub)\n\n## 3D重建 \u003Ca name=\"3DReconstruction\"\u002F>\n\n3D重建是指捕捉真实物体的形状和外观的过程。\n\n| 网络  | 大小（MB）| 位置|示例输出|\n| ------------- | ------------- |------------- |------------- |\n| [NeRF](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Fnerf)   | 3.78 |[GitHub](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Fnerf)|![NeRF](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_bed99f087597.jpg) |\n\n[返回顶部](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#matlab-deep-learning-model-hub)\n\n## 视频分类 \u003Ca name=\"VideoClassification\"\u002F>\n\n视频分类是一种计算机视觉技术，用于对一系列视频帧中的动作或内容进行分类。\n\n所有输入仅为视频或包含光流数据的视频，输出为手势分类及其置信度分数。\n\n![](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_8213a60b95e2.png)\n\n| 网络  | 输入 | 大小(MB) | 分类（人体动作）| 描述 | 位置|\n| ------------- | ------------- |------------- |------------- |------------- |------------- |\n| [SlowFast](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fref\u002Fslowfastvideoclassifier.html) | 视频 | 124 |400 | 比膨胀3D网络收敛更快 |[文档](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fref\u002Fslowfastvideoclassifier.html)\n| [R(2+1)D](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fref\u002Fr2plus1dvideoclassifier.html) | 视频 | 112 |400 | 比膨胀3D网络收敛更快|[文档](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fref\u002Fr2plus1dvideoclassifier.html)\n| [膨胀3D](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fref\u002Finflated3dvideoclassifier.html) | 视频及光流数据 | 91 | 400 | 当结合光流和RGB数据时，分类器的准确率会提高。| [文档](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fref\u002Finflated3dvideoclassifier.html)\n\n[返回顶部](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#matlab-deep-learning-model-hub)\n\n## 文本检测与识别 \u003Ca name=\"textdetection\"\u002F>\n\n文本检测是一种计算机视觉技术，用于在图像中定位文本实例。\n\n输入为RGB图像，输出是标识文本区域的边界框。\n\n![](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_5d0127dedf7d.png)\n\n| 网络  | 应用 | 大小（MB） | 位置 |\n| ------------- | ------------- |------------- |------------- |\n| [CRAFT](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fref\u002Fdetecttextcraft.html)   | 经过训练可检测英语、韩语、意大利语、法语、阿拉伯语、德语和孟加拉语（印度）。| 3.8 |[文档](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fref\u002Fdetecttextcraft.html) \u003Cbr \u002F> [GitHub](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FText-Detection-using-Deep-Learning) |\n\n**特定应用的文本检测器**\n\n| 网络  | 应用 | 大小（MB） | 位置 |示例输出 |\n| ------------- | ------------- |------------- |------------- |------------- |\n| [七段数码管识别](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FSeven-Segment-Digit-Recognition)   | 使用深度学习和OCR技术进行七段数码管识别。这在工业自动化应用中非常有用，尤其是在数字显示屏周围背景复杂的情况下。 | 3.8 |[文档](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fref\u002Focr.html) \u003Cbr \u002F> [GitHub](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FSeven-Segment-Digit-Recognition)  |![](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_2e64190ead1d.png)|\n\n[返回顶部](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#matlab-deep-learning-model-hub)\n\n## 变压器（文本） \u003Ca name=\"transformers\"\u002F>\n\n预训练的Transformer模型已经学会了从文本中提取强大且信息丰富的特征。可以将其作为起点，利用迁移学习来完成新的任务。\n\n输入为文本序列，输出为文本特征嵌入。\n\n![](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_2502418d7649.png)\n\n\n| 网络  | 应用 | 大小（MB） | 位置 |\n| ------------- | ------------- |------------ |------------ |\n| [BERT](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Ftextanalytics\u002Fref\u002Fbert.html)   | 特征提取（句子和词嵌入）、文本分类、标记分类、掩码语言建模、问答 |390 |[GitHub](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Ftransformer-models#bert-and-finbert) \u003Cbr \u002F> [文档](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Ftextanalytics\u002Fref\u002Fbert.html) | \n| [all-MiniLM-L6-v2](https:\u002F\u002Fwww.mathworks.com\u002Fmatlabcentral\u002Ffileexchange\u002F156399-text-analytics-toolbox-model-for-all-minilm-l6-v2-network)   | 文档嵌入、聚类、信息检索  |80 |[文档](https:\u002F\u002Fwww.mathworks.com\u002Fmatlabcentral\u002Ffileexchange\u002F156399-text-analytics-toolbox-model-for-all-minilm-l6-v2-network)  | \n| [all-MiniLM-L12-v2](https:\u002F\u002Fwww.mathworks.com\u002Fmatlabcentral\u002Ffileexchange\u002F156394-text-analytics-toolbox-model-for-all-minilm-l12-v2-network)   | 文档嵌入、聚类、信息检索  |120 |[文档](https:\u002F\u002Fwww.mathworks.com\u002Fmatlabcentral\u002Ffileexchange\u002F156394-text-analytics-toolbox-model-for-all-minilm-l12-v2-network)  | \n\n**特定应用的变压器**\n\n| 网络  | 应用 | 大小（MB） | 位置 | 输出示例 |\n| ------------- | ------------- | ------------- |------------- |------------- |\n| [FinBERT](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Ftransformer-models#bert-and-finbert)   | FinBERT模型是一种用于金融情感分析的BERT模型 | 388 |[GitHub](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Ftransformer-models#bert-and-finbert) |![](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_6b916896ae8a.png)|\n| [GPT-2](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Ftransformer-models#gpt-2) | GPT-2模型是一个用于文本摘要的解码器模型。| 1.2GB |[GitHub](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Ftransformer-models#gpt-2) |![](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_0694504b0d56.png)|\n\n[返回顶部](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#matlab-deep-learning-model-hub)\n\n## 音频嵌入 \u003Ca name=\"AudioEmbeddings\"\u002F>\n\n音频嵌入的预训练模型已经学会了从音频信号中提取强大且信息丰富的特征。可以将其作为起点，利用迁移学习来完成新的任务。\n\n输入为音频信号，输出为音频特征嵌入。\n\n![](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_32113253007f.png)\n\n**注2：自R2024a版本起，请使用[audiopretrainedNetwork](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Faudio\u002Fref\u002Faudiopretrainednetwork.html)函数，并指定预训练模型**。例如，要访问VGGish，可以使用以下代码：\n```\nnet = audioPretrainedNetwork(\"vggish\");\n```\n\n| 网络  | 应用 | 大小（MB） | 位置 |\n| ------------- | ------------- | ------------- |------------- |\n| [VGGish](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Faudio\u002Fref\u002Fvggish.html)\u003Csup>2\u003Csup>   | 特征嵌入 | 257 |[文档](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Faudio\u002Fref\u002Fvggish.html) |\n| [OpenL3](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Faudio\u002Fref\u002Fopenl3.html)\u003Csup>2\u003Csup>   | 特征嵌入 | 200 |[文档](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Faudio\u002Fref\u002Fopenl3.html) |\n\n## 音频专用模型\u003Ca name=\"Application Specific Audio Models\"\u002F>\n\n| 网络  | 应用 | 大小 (MB) | 输出类别 | 位置 | 输出示例 |\n| ------------- | ------------- | ------------- |------------- |------------- |------------- |\n| \u003Ca name=\"SoundClassification\"\u002F>[vadnet](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Faudio\u002Fref\u002Fvadnet.html)\u003Csup>2\u003Csup>   | 语音活动检测（回归） | 0.427 | - |[文档](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Faudio\u002Fref\u002Fvadnet.html) |\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_4e71efcc5aa3.png\" width=150>|\n| \u003Ca name=\"SoundClassification\"\u002F>[YAMNet](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Faudio\u002Fref\u002Fyamnet.html)\u003Csup>2\u003Csup>   | 声音分类 | 13.5 | 521 |[文档](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Faudio\u002Fref\u002Fyamnet.html) |\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_5fb224370fce.png\" width=150>|\n| \u003Ca name=\"PitchEstimation\"\u002F>[CREPE](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Faudio\u002Fref\u002Fcrepe.html)\u003Csup>2\u003Csup>   | 音高估计（回归） | 132| - |[文档](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Faudio\u002Fref\u002Fcrepe.html) |\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_a90b8645ea33.png\" width=150>|\n\n## 语音转文本 \u003Ca name=\"Speech2Text\"\u002F>\n\n语音转文本模型提供了一种快速高效的将口语转换为书面文本的方法，这不仅增强了残疾人士的可访问性，还支持诸如文本摘要和情感分析等下游任务，并简化了文档处理流程。作为人机交互界面的关键组成部分，例如个人助理，它允许自然直观的交互，使机器能够理解并执行语音命令，从而提高易用性并在各种应用中实现更广泛的包容性。\n\n输入是音频信号，输出是文本。\n\n![](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_34a5ee07f3d2.png)\n\n| 网络  | 应用 | 大小 (MB) | 字错误率 (WER) | 位置 | \n| ------------- | ------------- | ------------- |------------- |------------- |\n| [wav2vec](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Fwav2vec-2.0)   | 语音转文本 | 236| 3.2 |[GitHub](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Fwav2vec-2.0) |\n| [deepspeech](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Fdeepspeech)   | 语音转文本 | 167| 5.97 |[GitHub](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Fdeepspeech) |\n\n[返回顶部](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#matlab-deep-learning-model-hub)\n\n## 激光雷达 \u003Ca name=\"PointCloud\"\u002F>\n\n点云数据可通过多种传感器获取，如激光雷达、雷达和深度相机。由于每个物体的数据稀疏、物体遮挡以及传感器噪声等问题，使用点云数据训练鲁棒的分类器颇具挑战性。而深度学习技术已被证明能够直接从点云数据中学习鲁棒的特征表示，从而有效应对这些挑战。\n\n输入是转换为五通道的激光雷达点云，输出是在点云上叠加的分割、分类或目标检测结果。\n\n![](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_bbd74a3633b7.png)\n\n| 网络  | 应用 | 大小 (MB) | 物体类别 | 位置 | \n| ------------- | ------------- | ------------- |------------- |------------- |\n| [PointNet](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fug\u002Fpoint-cloud-classification-using-pointnet-deep-learning.html)  | 分类 | 5| 14 |[文档](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fug\u002Fpoint-cloud-classification-using-pointnet-deep-learning.html)|\n| \u003Ca name=\"PointCloudSeg\"\u002F>[PointNet++](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Flidar\u002Fug\u002Faerial-lidar-segmentation-using-pointnet-network.html)  | 分割 | 3| 8 |[文档](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Flidar\u002Fug\u002Faerial-lidar-segmentation-using-pointnet-network.html)|\n| [PointSeg](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fug\u002Flidar-semantic-segmentation-using-pointseg.html)   | 分割 | 14| 3 |[文档](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fvision\u002Fug\u002Fpoint-cloud-classification-using-pointnet-deep-learning.html)|\n| [SqueezeSegV2](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fug\u002Flidar-semantic-segmentation-using-squeezesegv2.html)   | 分割 |5| 12 |[文档](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fug\u002Flidar-semantic-segmentation-using-squeezesegv2.html) |\n| [SalsaNext](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Fpretrained-salsanext)   | 分割 |20.9 | 13 |[GitHub](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Fpretrained-salsanext)|\n| \u003Ca name=\"PointCloudObj\"\u002F>[PointPillars](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Flidar\u002Fug\u002Fobject-detection-using-pointpillars-network.html)   | 目标检测 | 8| 3 |[文档](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Flidar\u002Fug\u002Fobject-detection-using-pointpillars-network.html)|\n| [Complex YOLO v4](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FLidar-object-detection-using-complex-yolov4)   | 目标检测 | 233（complex-yolov4） \u003Cbr \u002F> 21（tiny-complex-yolov4） | 3 |[GitHub](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FLidar-object-detection-using-complex-yolov4)|\n\n[返回顶部](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#matlab-deep-learning-model-hub)\n\n## 机械臂运动规划 \u003Ca name=\"ManipMotionPlanning\"\u002F>\n\n机械臂运动规划是一种在障碍物环境中规划机器人手臂从起始位置到目标位置轨迹的技术。\n  \n预训练的深度学习模型已经学会了为重复性任务（如抓取和放置物体）规划此类轨迹，从而比传统算法显著提高了速度。\n\n输入是机器人的起始配置、目标配置以及障碍物环境编码，输出是中间轨迹的预测。\n\n![](Images\u002Fmotion_planning_workflow.svg)\n\n| 网络  | 应用 | 大小 (MB)| 位置|\n| --- | --- | --- | --- |\n| [基于深度学习的CHOMP (DLCHOMP)](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Freleases\u002FR2024a\u002Frobotics\u002Fref\u002Fdlchomp.html)   | 轨迹预测 | 25 | [文档](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Freleases\u002FR2024a\u002Frobotics\u002Fref\u002Fdlchomp.html)\u003Cbr \u002F>[GitHub](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002Fpretrained-dlchomp) |\n\n\n\n[返回顶部](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#matlab-deep-learning-model-hub)\n\n## 基于运动规划网络的路径规划 \u003Ca name=\"PathPlanningMPNet\"\u002F>\n\n运动规划网络（MPNet）是一种基于深度学习的方法，用于在运动规划问题中找到起点和目标点之间的最优路径。MPNet 是一种深度神经网络，可以在多个环境中进行训练，以学习环境中不同状态之间的最优路径。利用这些先验知识，MPNet 可以：\n- 在未知测试环境中生成两个状态之间的有信息量的采样点。这些采样点可以与基于采样的运动规划算法（如最优快速探索随机树 RRT*）结合使用，用于路径规划。\n- 在未知测试环境中计算两个状态之间的无碰撞路径。基于 MPNet 的路径规划器比传统的路径规划算法（如 RRT*）更加高效。\n\n欲了解更多信息，请访问 [开始使用运动规划网络](https:\u002F\u002Fin.mathworks.com\u002Fhelp\u002Fnav\u002Fug\u002Fget-started-with-motion-planning-networks.html)。\n\n![](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_readme_696259714deb.png)\n\n| 网络                                                                                                     | 应用   | 大小 (MB) | 位置                                                                                                |\n| ----------------------------------------------------------------------------------------------------------- | ------------- | --------- | ------------------------------------------------------------------------------------------------------- |\n| [mazeMapTrainedMPNET](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fnav\u002Fug\u002Fget-started-with-motion-planning-networks.html) | 运动规划 | 0.23      | [文档](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fnav\u002Fug\u002Ftrain-deep-learning-based-sampler-for-motion-planning.html) |\n\n[返回顶部](https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub#matlab-deep-learning-model-hub)\n\n## 模型请求\n如果您希望请求 MATLAB 对更多预训练模型的支持，[**请在此仓库中创建一个问题**](https:\u002F\u002Fdocs.github.com\u002Fen\u002Fissues\u002Ftracking-your-work-with-issues\u002Fcreating-an-issue)。\n\n或者，您也可以通过以下方式发送请求：\n\n[Jianghao Wang](https:\u002F\u002Fwww.linkedin.com\u002Fin\u002Fjianghao-wang-896aa1a4\u002F) \u003Cbr \u002F> \n深度学习产品经理 \u003Cbr \u002F> \njianghaw@mathworks.com\n\n版权所有 2024 年，The MathWorks, Inc.","# MATLAB 深度学习模型中心 (MATLAB Deep Learning Model Hub) 快速上手指南\n\n本指南帮助中国开发者快速在 MATLAB 中获取并使用预训练的深度学习模型，涵盖计算机视觉、自然语言处理、音频及机器人等领域。\n\n## 1. 环境准备\n\n在使用本工具前，请确保您的开发环境满足以下要求：\n\n*   **操作系统**: Windows, macOS, 或 Linux。\n*   **MATLAB 版本**: \n    *   推荐 **R2024a** 或更高版本（可使用最新的 `imagePretrainedNetwork` 函数）。\n    *   旧版本用户需使用特定的网络加载函数（如 `googlenet`, `resnet50` 等）。\n*   **必备工具箱 (Toolboxes)**:\n    *   **Deep Learning Toolbox**: 核心依赖，用于构建和训练深度学习网络。\n    *   **Computer Vision Toolbox**: 用于图像分类、目标检测等视觉任务。\n    *   **Text Analytics Toolbox**: 用于自然语言处理任务（如 Transformers）。\n    *   **Audio Toolbox**: 用于音频分类和语音识别任务。\n    *   **Lidar Toolbox** \u002F **Robotics System Toolbox**: 用于点云处理和机器人路径规划（按需安装）。\n\n> **注意**: 请确保已登录 MathWorks 账号，以便下载预训练模型权重。国内用户若遇到下载缓慢问题，建议检查网络连接或使用稳定的网络环境。\n\n## 2. 安装与获取模型\n\n本“模型中心”并非一个需要单独安装的第三方包，而是 MATLAB 官方内置模型库的索引集合。大多数主流模型（如 ResNet, YOLO, GoogLeNet）可直接通过 MATLAB 命令调用，系统会自动下载权重文件。\n\n对于部分托管在 GitHub 上的最新或特定变体模型（如 YOLOv9, ConvMixer），您需要克隆对应的仓库。\n\n### 方式 A：直接调用内置模型（推荐）\n无需手动安装，MATLAB 会在首次调用时自动下载模型。\n\n### 方式 B：获取 GitHub 托管模型\n如果您需要使用 README 中列出的特定 GitHub 模型（例如最新的 YOLOv9）：\n\n```bash\n# 示例：克隆 YOLOv9 预训练模型仓库\ngit clone https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FPretrained-Yolov9-Network-For-Object-Detection.git\n```\n*注：国内访问 GitHub 若受限，可配置本地 Git 代理或使用 Gitee 镜像（如有）。*\n\n## 3. 基本使用\n\n以下是基于 **R2024a** 及更新版本的最简使用示例。\n\n### 场景一：图像分类 (Image Classification)\n\n加载预训练的 **GoogLeNet** 模型并对图片进行分类。\n\n```matlab\n% 1. 加载预训练网络和类别标签\n% R2024a 及以上版本推荐使用 imagePretrainedNetwork\n[net, classes] = imagePretrainedNetwork(\"googlenet\");\n\n% 2. 读取并预处理图像\nimg = imread(\"example_image.jpg\");\nimg_resized = imresize(img, net.InputSize(1:2)); % 调整大小以匹配网络输入\n\n% 3. 执行预测\n[labelIdx, scores] = classify(net, img_resized);\nlabel = classes(labelIdx);\n\n% 4. 显示结果\ndisp([\"预测类别：\", label]);\ndisp([\"置信度分数：\", num2str(max(scores))]);\n```\n\n*支持的其他经典模型*: `\"resnet50\"`, `\"mobilenetv2\"`, `\"efficientnetb0\"`, `\"vgg16\"` 等，只需替换字符串参数即可。\n\n### 场景二：目标检测 (Object Detection)\n\n加载 **YOLO v8** 模型检测图像中的物体。\n\n```matlab\n% 1. 加载预训练的 YOLOv8 网络 (以 yolo8n 为例)\n% 确保已安装 Computer Vision Toolbox 并联网下载权重\ndetector = yolov8ObjectDetector(\"yolo8n\");\n\n% 2. 读取图像\nimg = imread(\"traffic_scene.jpg\");\n\n% 3. 执行检测\n[bboxes, labels, scores] = detect(detector, img);\n\n% 4. 可视化结果\noutputImg = insertObjectAnnotation(img, \"rectangle\", bboxes, labels);\nimshow(outputImg);\ntitle(\"Object Detection Results\");\n```\n\n*支持的检测模型*: `\"yolov4\"`, `\"yolox\"`, `\"efficientdet-d0\"` 等。\n\n### 场景三：迁移学习 (Transfer Learning)\n\n利用预训练模型作为起点，快速训练自定义任务（例如分类新的花卉品种）。\n\n```matlab\n% 1. 加载预训练网络 (例如 ResNet18)\n[net, info] = imagePretrainedNetwork(\"resnet18\");\n\n% 2. 准备自定义数据 (imagedatastore)\nimds = imageDatastore(\"my_custom_dataset\", \"IncludeSubfolders\", true, ...\n    \"LabelSource\", \"foldernames\");\n\n% 3. 修改最后一层以匹配新任务的类别数量\nnumClasses = numel(categories(imds.Labels));\nlgraph = layerGraph(net);\nlgraph = replaceLayer(lgraph, \"ClassificationLayer\", ...\n    classificationLayer(\"Name\", \"newClassLayer\", \"Classes\", numClasses));\n\n% 4. 设置训练选项并微调网络\noptions = trainingOptions(\"sgdm\", ...\n    \"InitialLearnRate\", 1e-4, ...\n    \"MaxEpochs\", 6, ...\n    \"MiniBatchSize\", 16, ...\n    \"Plots\", \"training-progress\");\n\ntrainedNet = trainNetwork(imds, lgraph, options);\n```\n\n---\n**提示**: \n*   选择模型时，请在**准确率 (Accuracy\u002FmAP)**、**速度 (Speed)** 和 **模型大小 (Size)** 之间进行权衡。移动端部署建议选择 MobileNet 或 YOLO-Nano 系列；高精度服务器端任务可选择 ResNet101 或 EfficientNet-B7。\n*   所有模型的详细文档可通过 MATLAB 命令行输入 `doc \u003C模型名称>` 查看（例如 `doc resnet50`）。","某汽车制造厂的算法工程师正致力于开发一套基于车载摄像头的实时交通标志识别系统，以辅助自动驾驶车辆安全行驶。\n\n### 没有 MATLAB-Deep-Learning-Model-Hub 时\n- **数据收集成本高昂**：团队需从零开始采集并标注数十万张不同光照、天气下的交通标志图片，耗时数月且人力成本巨大。\n- **模型训练周期漫长**：从头训练深度神经网络需要高性能 GPU 集群支持，且调参过程复杂，往往数周才能收敛到一个可用基准。\n- **小样本效果不佳**：由于罕见标志（如施工临时标牌）样本稀缺，从头训练的模型泛化能力差，误识率居高不下。\n- **工程落地门槛高**：缺乏经过验证的成熟架构，工程师需花费大量精力复现论文代码并解决兼容性问题，延误项目进度。\n\n### 使用 MATLAB-Deep-Learning-Model-Hub 后\n- **即刻获取预训练模型**：直接调用 Hub 中已在百万级图像上训练好的 GoogLeNet 或 SqueezeNet 模型，作为特征提取器立即启动开发。\n- **高效迁移学习**：仅需少量本地采集的交通标志图片进行微调（Transfer Learning），即可在普通工作站上于数小时内完成高精度模型定制。\n- **显著提升识别准确率**：利用预训练模型强大的通用特征提取能力，即使针对罕见标志也能实现稳健识别，大幅降低漏检率。\n- **无缝集成工作流**：通过 `imagePretrainedNetwork` 函数一行代码即可加载模型，无缝对接 MATLAB 现有的仿真与代码生成工具链，加速部署上车。\n\nMATLAB-Deep-Learning-Model-Hub 将原本需要数月的基础研发工作压缩至几天，让团队能专注于特定场景的业务逻辑优化而非重复造轮子。","https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fmatlab-deep-learning_MATLAB-Deep-Learning-Model-Hub_1096ab4c.png","matlab-deep-learning","MATLAB Deep Learning","https:\u002F\u002Foss.gittoolsai.com\u002Favatars\u002Fmatlab-deep-learning_3d43002c.png","Deep learning resources, including pretrained neural network models.",null,"https:\u002F\u002Fwww.mathworks.com\u002Fsolutions\u002Fdeep-learning.html","https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning",[84],{"name":85,"color":86,"percentage":87},"MATLAB","#e16737",100,554,117,"2026-04-03T01:12:40","NOASSERTION","未说明 (依赖 MATLAB 支持的平台，通常包括 Windows, macOS, Linux)","未说明 (取决于具体使用的模型和 MATLAB 版本，部分大模型如 Vision Transformer Large 可能需要高性能 GPU)","未说明 (取决于模型大小，大型模型如 VGG19\u002FInceptionResNetV2 需较大内存)",{"notes":96,"python":97,"dependencies":98},"该工具是 MATLAB 的预训练模型库，非独立 Python 项目。自 R2024a 起，建议使用内置的 imagePretrainedNetwork 函数加载模型。不同模型对硬件要求差异巨大：轻量级模型（如 SqueezeNet, 5.2MB）可在 CPU 运行，而大型模型（如 Vision Transformer Large, 1.1GB 或 NASNetLarge, 332MB）建议配备专用 GPU 以加速推理和训练。具体算子支持需参考对应 MATLAB 版本的文档。","不适用 (基于 MATLAB 环境)",[99,100,101,102],"MATLAB R2024a 或更高版本 (推荐使用 imagePretrainedNetwork 函数)","Deep Learning Toolbox","Computer Vision Toolbox","Image Processing Toolbox",[13],[105,106,107,76,108,109,110],"matlab","deep-learning","pretrained-models","model-zoo","awesome","awesome-list","2026-03-27T02:49:30.150509","2026-04-06T05:15:23.898064",[114,119,123],{"id":115,"question_zh":116,"answer_zh":117,"source_url":118},14468,"在哪里可以找到预训练的 PointNet++ 模型？","MathWorks Model Hub 中列出了预训练的 PointNet++ 模型，具体应用于“航空激光雷达语义分割”（Aerial Lidar Semantic Segmentation）。您可以通过以下链接访问示例并加载模型：\nhttps:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Flidar\u002Fug\u002Faerial-lidar-segmentation-using-pointnet-network.html\n\n在 MATLAB 中加载预训练模型的命令为：\n>> load('pointnetplusTrained','net');","https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub\u002Fissues\u002F4",{"id":120,"question_zh":121,"answer_zh":122,"source_url":118},14469,"如果 PointNet++ 模型训练无法收敛，应该怎么办？","首先尝试使用官方提供的预训练模型进行迁移学习。您可以加载航空激光雷达语义分割示例中的预训练模型（命令：load('pointnetplusTrained','net');），然后针对您的特定任务进行调整。如果问题仍然存在，建议分享您正在使用的代码和数据链接，以便获得更具体的帮助。",{"id":124,"question_zh":125,"answer_zh":126,"source_url":127},14470,"MATLAB 是否计划支持 Vision Transformer (ViT) 模型？","是的，开发团队正在研究添加对 Vision Transformers (ViT) 的支持。一旦有更新进展，相关公告会在该项目的 Issue 页面或 Model Hub 中发布。","https:\u002F\u002Fgithub.com\u002Fmatlab-deep-learning\u002FMATLAB-Deep-Learning-Model-Hub\u002Fissues\u002F1",[129,134,139,143],{"id":130,"version":131,"summary_zh":132,"released_at":133},81385,"1.0.3","自 MATLAB R2024a 起，新增了用于加载随产品提供的预训练模型的推荐函数。 有关详细信息，请参阅相关文档页面（[imagePretrainedNetwork](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Fdeeplearning\u002Fref\u002Fimagepretrainednetwork.html) 和 [audioPretrainedNetwork](https:\u002F\u002Fwww.mathworks.com\u002Fhelp\u002Faudio\u002Fref\u002Faudiopretrainednetwork.html)）。","2024-03-27T21:08:39",{"id":135,"version":136,"summary_zh":137,"released_at":138},81386,"1.0.2","针对23a更新","2023-05-30T11:36:07",{"id":140,"version":141,"summary_zh":80,"released_at":142},81387,"1.0.1","2022-03-15T13:02:12",{"id":144,"version":145,"summary_zh":80,"released_at":146},81388,"1.0","2021-12-16T14:26:41"]