[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"similar-leggedrobotics--darknet_ros":3,"tool-leggedrobotics--darknet_ros":64},[4,17,27,35,43,56],{"id":5,"name":6,"github_repo":7,"description_zh":8,"stars":9,"difficulty_score":10,"last_commit_at":11,"category_tags":12,"status":16},3808,"stable-diffusion-webui","AUTOMATIC1111\u002Fstable-diffusion-webui","stable-diffusion-webui 是一个基于 Gradio 构建的网页版操作界面，旨在让用户能够轻松地在本地运行和使用强大的 Stable Diffusion 图像生成模型。它解决了原始模型依赖命令行、操作门槛高且功能分散的痛点，将复杂的 AI 绘图流程整合进一个直观易用的图形化平台。\n\n无论是希望快速上手的普通创作者、需要精细控制画面细节的设计师，还是想要深入探索模型潜力的开发者与研究人员，都能从中获益。其核心亮点在于极高的功能丰富度：不仅支持文生图、图生图、局部重绘（Inpainting）和外绘（Outpainting）等基础模式，还独创了注意力机制调整、提示词矩阵、负向提示词以及“高清修复”等高级功能。此外，它内置了 GFPGAN 和 CodeFormer 等人脸修复工具，支持多种神经网络放大算法，并允许用户通过插件系统无限扩展能力。即使是显存有限的设备，stable-diffusion-webui 也提供了相应的优化选项，让高质量的 AI 艺术创作变得触手可及。",162132,3,"2026-04-05T11:01:52",[13,14,15],"开发框架","图像","Agent","ready",{"id":18,"name":19,"github_repo":20,"description_zh":21,"stars":22,"difficulty_score":23,"last_commit_at":24,"category_tags":25,"status":16},1381,"everything-claude-code","affaan-m\u002Feverything-claude-code","everything-claude-code 是一套专为 AI 编程助手（如 Claude Code、Codex、Cursor 等）打造的高性能优化系统。它不仅仅是一组配置文件，而是一个经过长期实战打磨的完整框架，旨在解决 AI 代理在实际开发中面临的效率低下、记忆丢失、安全隐患及缺乏持续学习能力等核心痛点。\n\n通过引入技能模块化、直觉增强、记忆持久化机制以及内置的安全扫描功能，everything-claude-code 能显著提升 AI 在复杂任务中的表现，帮助开发者构建更稳定、更智能的生产级 AI 代理。其独特的“研究优先”开发理念和针对 Token 消耗的优化策略，使得模型响应更快、成本更低，同时有效防御潜在的攻击向量。\n\n这套工具特别适合软件开发者、AI 研究人员以及希望深度定制 AI 工作流的技术团队使用。无论您是在构建大型代码库，还是需要 AI 协助进行安全审计与自动化测试，everything-claude-code 都能提供强大的底层支持。作为一个曾荣获 Anthropic 黑客大奖的开源项目，它融合了多语言支持与丰富的实战钩子（hooks），让 AI 真正成长为懂上",138956,2,"2026-04-05T11:33:21",[13,15,26],"语言模型",{"id":28,"name":29,"github_repo":30,"description_zh":31,"stars":32,"difficulty_score":23,"last_commit_at":33,"category_tags":34,"status":16},2271,"ComfyUI","Comfy-Org\u002FComfyUI","ComfyUI 是一款功能强大且高度模块化的视觉 AI 引擎，专为设计和执行复杂的 Stable Diffusion 图像生成流程而打造。它摒弃了传统的代码编写模式，采用直观的节点式流程图界面，让用户通过连接不同的功能模块即可构建个性化的生成管线。\n\n这一设计巧妙解决了高级 AI 绘图工作流配置复杂、灵活性不足的痛点。用户无需具备编程背景，也能自由组合模型、调整参数并实时预览效果，轻松实现从基础文生图到多步骤高清修复等各类复杂任务。ComfyUI 拥有极佳的兼容性，不仅支持 Windows、macOS 和 Linux 全平台，还广泛适配 NVIDIA、AMD、Intel 及苹果 Silicon 等多种硬件架构，并率先支持 SDXL、Flux、SD3 等前沿模型。\n\n无论是希望深入探索算法潜力的研究人员和开发者，还是追求极致创作自由度的设计师与资深 AI 绘画爱好者，ComfyUI 都能提供强大的支持。其独特的模块化架构允许社区不断扩展新功能，使其成为当前最灵活、生态最丰富的开源扩散模型工具之一，帮助用户将创意高效转化为现实。",107662,"2026-04-03T11:11:01",[13,14,15],{"id":36,"name":37,"github_repo":38,"description_zh":39,"stars":40,"difficulty_score":23,"last_commit_at":41,"category_tags":42,"status":16},3704,"NextChat","ChatGPTNextWeb\u002FNextChat","NextChat 是一款轻量且极速的 AI 助手，旨在为用户提供流畅、跨平台的大模型交互体验。它完美解决了用户在多设备间切换时难以保持对话连续性，以及面对众多 AI 模型不知如何统一管理的痛点。无论是日常办公、学习辅助还是创意激发，NextChat 都能让用户随时随地通过网页、iOS、Android、Windows、MacOS 或 Linux 端无缝接入智能服务。\n\n这款工具非常适合普通用户、学生、职场人士以及需要私有化部署的企业团队使用。对于开发者而言，它也提供了便捷的自托管方案，支持一键部署到 Vercel 或 Zeabur 等平台。\n\nNextChat 的核心亮点在于其广泛的模型兼容性，原生支持 Claude、DeepSeek、GPT-4 及 Gemini Pro 等主流大模型，让用户在一个界面即可自由切换不同 AI 能力。此外，它还率先支持 MCP（Model Context Protocol）协议，增强了上下文处理能力。针对企业用户，NextChat 提供专业版解决方案，具备品牌定制、细粒度权限控制、内部知识库整合及安全审计等功能，满足公司对数据隐私和个性化管理的高标准要求。",87618,"2026-04-05T07:20:52",[13,26],{"id":44,"name":45,"github_repo":46,"description_zh":47,"stars":48,"difficulty_score":23,"last_commit_at":49,"category_tags":50,"status":16},2268,"ML-For-Beginners","microsoft\u002FML-For-Beginners","ML-For-Beginners 是由微软推出的一套系统化机器学习入门课程，旨在帮助零基础用户轻松掌握经典机器学习知识。这套课程将学习路径规划为 12 周，包含 26 节精炼课程和 52 道配套测验，内容涵盖从基础概念到实际应用的完整流程，有效解决了初学者面对庞大知识体系时无从下手、缺乏结构化指导的痛点。\n\n无论是希望转型的开发者、需要补充算法背景的研究人员，还是对人工智能充满好奇的普通爱好者，都能从中受益。课程不仅提供了清晰的理论讲解，还强调动手实践，让用户在循序渐进中建立扎实的技能基础。其独特的亮点在于强大的多语言支持，通过自动化机制提供了包括简体中文在内的 50 多种语言版本，极大地降低了全球不同背景用户的学习门槛。此外，项目采用开源协作模式，社区活跃且内容持续更新，确保学习者能获取前沿且准确的技术资讯。如果你正寻找一条清晰、友好且专业的机器学习入门之路，ML-For-Beginners 将是理想的起点。",84991,"2026-04-05T10:45:23",[14,51,52,53,15,54,26,13,55],"数据工具","视频","插件","其他","音频",{"id":57,"name":58,"github_repo":59,"description_zh":60,"stars":61,"difficulty_score":10,"last_commit_at":62,"category_tags":63,"status":16},3128,"ragflow","infiniflow\u002Fragflow","RAGFlow 是一款领先的开源检索增强生成（RAG）引擎，旨在为大语言模型构建更精准、可靠的上下文层。它巧妙地将前沿的 RAG 技术与智能体（Agent）能力相结合，不仅支持从各类文档中高效提取知识，还能让模型基于这些知识进行逻辑推理和任务执行。\n\n在大模型应用中，幻觉问题和知识滞后是常见痛点。RAGFlow 通过深度解析复杂文档结构（如表格、图表及混合排版），显著提升了信息检索的准确度，从而有效减少模型“胡编乱造”的现象，确保回答既有据可依又具备时效性。其内置的智能体机制更进一步，使系统不仅能回答问题，还能自主规划步骤解决复杂问题。\n\n这款工具特别适合开发者、企业技术团队以及 AI 研究人员使用。无论是希望快速搭建私有知识库问答系统，还是致力于探索大模型在垂直领域落地的创新者，都能从中受益。RAGFlow 提供了可视化的工作流编排界面和灵活的 API 接口，既降低了非算法背景用户的上手门槛，也满足了专业开发者对系统深度定制的需求。作为基于 Apache 2.0 协议开源的项目，它正成为连接通用大模型与行业专有知识之间的重要桥梁。",77062,"2026-04-04T04:44:48",[15,14,13,26,54],{"id":65,"github_repo":66,"name":67,"description_en":68,"description_zh":69,"ai_summary_zh":69,"readme_en":70,"readme_zh":71,"quickstart_zh":72,"use_case_zh":73,"hero_image_url":74,"owner_login":75,"owner_name":76,"owner_avatar_url":77,"owner_bio":78,"owner_company":79,"owner_location":79,"owner_email":79,"owner_twitter":75,"owner_website":80,"owner_url":81,"languages":82,"stars":91,"forks":92,"last_commit_at":93,"license":94,"difficulty_score":95,"env_os":96,"env_gpu":97,"env_ram":98,"env_deps":99,"category_tags":105,"github_topics":106,"view_count":23,"oss_zip_url":79,"oss_zip_packed_at":79,"status":16,"created_at":115,"updated_at":116,"faqs":117,"releases":158},1363,"leggedrobotics\u002Fdarknet_ros","darknet_ros","YOLO ROS: Real-Time Object Detection for ROS","darknet_ros 把大名鼎鼎的 YOLO 目标检测算法无缝搬进 ROS 生态，让机器人相机画面里的行人、车辆、动物、家具等 80 类常见物体被实时框选出来。它解决了传统视觉节点检测慢、类别少、配置繁琐的问题，只需一条 ROS 话题即可拿到带边框的图像和物体坐标，方便做导航、抓取或安防。  \n支持 ROS Noetic\u002FMelodic 及 ROS2，GPU 与 CPU 都能跑，也允许你替换自己的训练权重。适合机器人开发者、SLAM 研究者或任何想在 ROS 里快速获得“看见世界”能力的团队。","# YOLO ROS: Real-Time Object Detection for ROS\n\n## Overview\n\nThis is a ROS package developed for **object detection in camera images**. You only look once (YOLO) is a state-of-the-art, real-time object detection system. In the following ROS package you are able to use **YOLO (V3) on GPU and CPU**. The pre-trained model of the convolutional neural network is able to detect pre-trained classes including the data set from VOC and COCO, or you can also create a network with your own detection objects. For more information about YOLO, Darknet, available training data and training YOLO see the following link: [YOLO: Real-Time Object Detection](http:\u002F\u002Fpjreddie.com\u002Fdarknet\u002Fyolo\u002F).\n\nThe YOLO packages have been tested under **ROS Noetic** and **Ubuntu 20.04**. Note: We also provide branches that work under **ROS Melodic**, **ROS Foxy** and **ROS2**.\n\nThis is research code, expect that it changes often and any fitness for a particular purpose is disclaimed.\n\n**Author: [Marko Bjelonic](https:\u002F\u002Fwww.markobjelonic.com), marko.bjelonic@mavt.ethz.ch**\n\n**Affiliation: [Robotic Systems Lab](http:\u002F\u002Fwww.rsl.ethz.ch\u002F), ETH Zurich**\n\n![Darknet Ros example: Detection image](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fleggedrobotics_darknet_ros_readme_0d5c1b7a144c.png)\n![Darknet Ros example: Detection image](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fleggedrobotics_darknet_ros_readme_22d29d505435.png)\n\nBased on the [Pascal VOC](https:\u002F\u002Fpjreddie.com\u002Fprojects\u002Fpascal-voc-dataset-mirror\u002F) 2012 dataset, YOLO can detect the 20 Pascal object classes:\n\n- person\n- bird, cat, cow, dog, horse, sheep\n- aeroplane, bicycle, boat, bus, car, motorbike, train\n- bottle, chair, dining table, potted plant, sofa, tv\u002Fmonitor\n\nBased on the [COCO](http:\u002F\u002Fcocodataset.org\u002F#home) dataset, YOLO can detect the 80 COCO object classes:\n\n- person\n- bicycle, car, motorbike, aeroplane, bus, train, truck, boat\n- traffic light, fire hydrant, stop sign, parking meter, bench\n- cat, dog, horse, sheep, cow, elephant, bear, zebra, giraffe\n- backpack, umbrella, handbag, tie, suitcase, frisbee, skis, snowboard, sports ball, kite, baseball bat, baseball glove, skateboard, surfboard, tennis racket\n- bottle, wine glass, cup, fork, knife, spoon, bowl\n- banana, apple, sandwich, orange, broccoli, carrot, hot dog, pizza, donut, cake\n- chair, sofa, pottedplant, bed, diningtable, toilet, tvmonitor, laptop, mouse, remote, keyboard, cell phone, microwave, oven, toaster, sink, refrigerator, book, clock, vase, scissors, teddy bear, hair drier, toothbrush\n\n## Citing\n\nThe YOLO methods used in this software are described in the paper: [You Only Look Once: Unified, Real-Time Object Detection](https:\u002F\u002Farxiv.org\u002Fabs\u002F1506.02640).\n\nIf you are using YOLO V3 for ROS, please add the following citation to your publication:\n\nM. Bjelonic\n**\"YOLO ROS: Real-Time Object Detection for ROS\"**,\nURL: https:\u002F\u002Fgithub.com\u002Fleggedrobotics\u002Fdarknet_ros, 2018.\n\n    @misc{bjelonicYolo2018,\n      author = {Marko Bjelonic},\n      title = {{YOLO ROS}: Real-Time Object Detection for {ROS}},\n      howpublished = {\\url{https:\u002F\u002Fgithub.com\u002Fleggedrobotics\u002Fdarknet_ros}},\n      year = {2016--2018},\n    }\n\n## Installation\n\n### Dependencies\n\nThis software is built on the Robotic Operating System ([ROS]), which needs to be [installed](http:\u002F\u002Fwiki.ros.org) first. Additionally, YOLO for ROS depends on following software:\n\n- [OpenCV](http:\u002F\u002Fopencv.org\u002F) (computer vision library),\n- [boost](http:\u002F\u002Fwww.boost.org\u002F) (c++ library),\n\n### Building\n\n[![Build Status](https:\u002F\u002Fci.leggedrobotics.com\u002FbuildStatus\u002Ficon?job=github_leggedrobotics\u002Fdarknet_ros\u002Fmaster)](https:\u002F\u002Fci.leggedrobotics.com\u002Fjob\u002Fgithub_leggedrobotics\u002Fjob\u002Fdarknet_ros\u002Fjob\u002Fmaster\u002F)\n\nIn order to install darknet_ros, clone the latest version using SSH (see [how to set up an SSH key](https:\u002F\u002Fconfluence.atlassian.com\u002Fbitbucket\u002Fset-up-an-ssh-key-728138079.html)) from this repository into your catkin workspace and compile the package using ROS.\n\n    cd catkin_workspace\u002Fsrc\n    git clone --recursive git@github.com:leggedrobotics\u002Fdarknet_ros.git\n    cd ..\u002F\n\nTo maximize performance, make sure to build in *Release* mode. You can specify the build type by setting\n\n    catkin_make -DCMAKE_BUILD_TYPE=Release\n\nor using the [Catkin Command Line Tools](http:\u002F\u002Fcatkin-tools.readthedocs.io\u002Fen\u002Flatest\u002Findex.html#)\n\n    catkin build darknet_ros -DCMAKE_BUILD_TYPE=Release\n\nDarknet on the CPU is fast (approximately 1.5 seconds on an Intel Core i7-6700HQ CPU @ 2.60GHz × 8) but it's like 500 times faster on GPU! You'll have to have an Nvidia GPU and you'll have to install CUDA. The CMakeLists.txt file automatically detects if you have CUDA installed or not. CUDA is a parallel computing platform and application programming interface (API) model created by Nvidia. If you do not have CUDA on your System the build process will switch to the CPU version of YOLO. If you are compiling with CUDA, you might receive the following build error:\n\n    nvcc fatal : Unsupported gpu architecture 'compute_61'.\n\nThis means that you need to check the compute capability (version) of your GPU. You can find a list of supported GPUs in CUDA here: [CUDA - WIKIPEDIA](https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FCUDA#Supported_GPUs). Simply find the compute capability of your GPU and add it into darknet_ros\u002FCMakeLists.txt. Simply add a similar line like\n\n    -O3 -gencode arch=compute_62,code=sm_62\n\n### Download weights\n\nThe yolo-voc.weights and tiny-yolo-voc.weights are downloaded automatically in the CMakeLists.txt file. If you need to download them again, go into the weights folder and download the two pre-trained weights from the COCO data set:\n\n    cd catkin_workspace\u002Fsrc\u002Fdarknet_ros\u002Fdarknet_ros\u002Fyolo_network_config\u002Fweights\u002F\n    wget http:\u002F\u002Fpjreddie.com\u002Fmedia\u002Ffiles\u002Fyolov2.weights\n    wget http:\u002F\u002Fpjreddie.com\u002Fmedia\u002Ffiles\u002Fyolov2-tiny.weights\n\nAnd weights from the VOC data set can be found here:\n\n    wget http:\u002F\u002Fpjreddie.com\u002Fmedia\u002Ffiles\u002Fyolov2-voc.weights\n    wget http:\u002F\u002Fpjreddie.com\u002Fmedia\u002Ffiles\u002Fyolov2-tiny-voc.weights\n\nAnd the pre-trained weight from YOLO v3 can be found here:\n\n    wget http:\u002F\u002Fpjreddie.com\u002Fmedia\u002Ffiles\u002Fyolov3-tiny.weights\n    wget http:\u002F\u002Fpjreddie.com\u002Fmedia\u002Ffiles\u002Fyolov3.weights\n\nThere are more pre-trained weights from different data sets reported [here](https:\u002F\u002Fpjreddie.com\u002Fdarknet\u002Fyolo\u002F).\n\n### Use your own detection objects\n\nIn order to use your own detection objects you need to provide your weights and your cfg file inside the directories:\n\n    catkin_workspace\u002Fsrc\u002Fdarknet_ros\u002Fdarknet_ros\u002Fyolo_network_config\u002Fweights\u002F\n    catkin_workspace\u002Fsrc\u002Fdarknet_ros\u002Fdarknet_ros\u002Fyolo_network_config\u002Fcfg\u002F\n\nIn addition, you need to create your config file for ROS where you define the names of the detection objects. You need to include it inside:\n\n    catkin_workspace\u002Fsrc\u002Fdarknet_ros\u002Fdarknet_ros\u002Fconfig\u002F\n\nThen in the launch file you have to point to your new config file in the line:\n\n    \u003Crosparam command=\"load\" ns=\"darknet_ros\" file=\"$(find darknet_ros)\u002Fconfig\u002Fyour_config_file.yaml\"\u002F>\n\n### Unit Tests\n\nRun the unit tests using the [Catkin Command Line Tools](http:\u002F\u002Fcatkin-tools.readthedocs.io\u002Fen\u002Flatest\u002Findex.html#)\n\n    catkin build darknet_ros --no-deps --verbose --catkin-make-args run_tests\n\nYou will see the image above popping up.\n\n## Basic Usage\n\nIn order to get YOLO ROS: Real-Time Object Detection for ROS to run with your robot, you will need to adapt a few parameters. It is the easiest if duplicate and adapt all the parameter files that you need to change from the `darknet_ros` package. These are specifically the parameter files in `config` and the launch file from the `launch` folder.\n\n## Nodes\n\n### Node: darknet_ros\n\nThis is the main YOLO ROS: Real-Time Object Detection for ROS node. It uses the camera measurements to detect pre-learned objects in the frames.\n\n### ROS related parameters\n\nYou can change the names and other parameters of the publishers, subscribers and actions inside `darknet_ros\u002Fconfig\u002Fros.yaml`.\n\n#### Subscribed Topics\n\n* **`\u002Fcamera_reading`** ([sensor_msgs\u002FImage])\n\n    The camera measurements.\n\n#### Published Topics\n\n* **`object_detector`** ([std_msgs::Int8])\n\n    Publishes the number of detected objects.\n\n* **`bounding_boxes`** ([darknet_ros_msgs::BoundingBoxes])\n\n    Publishes an array of bounding boxes that gives information of the position and size of the bounding box in pixel coordinates.\n\n* **`detection_image`** ([sensor_msgs::Image])\n\n    Publishes an image of the detection image including the bounding boxes.\n\n#### Actions\n\n* **`camera_reading`** ([sensor_msgs::Image])\n\n    Sends an action with an image and the result is an array of bounding boxes.\n\n### Detection related parameters\n\nYou can change the parameters that are related to the detection by adding a new config file that looks similar to `darknet_ros\u002Fconfig\u002Fyolo.yaml`.\n\n* **`image_view\u002Fenable_opencv`** (bool)\n\n    Enable or disable the open cv view of the detection image including the bounding boxes.\n\n* **`image_view\u002Fwait_key_delay`** (int)\n\n    Wait key delay in ms of the open cv window.\n\n* **`yolo_model\u002Fconfig_file\u002Fname`** (string)\n\n    Name of the cfg file of the network that is used for detection. The code searches for this name inside `darknet_ros\u002Fyolo_network_config\u002Fcfg\u002F`.\n\n* **`yolo_model\u002Fweight_file\u002Fname`** (string)\n\n    Name of the weights file of the network that is used for detection. The code searches for this name inside `darknet_ros\u002Fyolo_network_config\u002Fweights\u002F`.\n\n* **`yolo_model\u002Fthreshold\u002Fvalue`** (float)\n\n    Threshold of the detection algorithm. It is defined between 0 and 1.\n\n* **`yolo_model\u002Fdetection_classes\u002Fnames`** (array of strings)\n\n    Detection names of the network used by the cfg and weights file inside `darknet_ros\u002Fyolo_network_config\u002F`.\n","# YOLO ROS：面向 ROS 的实时目标检测\n\n## 概述\n\n本软件是一个专为**相机图像中的目标检测**而开发的 ROS 包。YOLO 是一款业界领先的实时目标检测系统，只需看一次即可（YOLO）。在本 ROS 包中，您可选择在 GPU 和 CPU 上使用 **YOLO (V3)**。该卷积神经网络的预训练模型能够检测包括 VOC 和 COCO 数据集在内的多种预训练类别；您也可以根据自己的检测对象自定义构建网络。如需了解更多关于 YOLO、Darknet 以及可用训练数据和 YOLO 训练过程的信息，请访问以下链接：[YOLO：实时目标检测](http:\u002F\u002Fpjreddie.com\u002Fdarknet\u002Fyolo\u002F)。\n\n本 YOLO 软件包已在 **ROS Noetic** 和 **Ubuntu 20.04** 环境下进行过测试。请注意：我们还提供了适用于 **ROS Melodic**、**ROS Foxy** 以及 **ROS2** 的分支版本。\n\n本代码为研究性代码，可能频繁更新，且不保证其适用于特定用途。\n\n**作者：Marko Bjelonic**（https:\u002F\u002Fwww.markobjelonic.com），marko.bjelonic@mavt.ethz.ch\n\n**所属机构：ETH Zurich 机器人系统实验室**（http:\u002F\u002Fwww.rsl.ethz.ch\u002F）\n\n![Darknet ROS 示例：检测图像](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fleggedrobotics_darknet_ros_readme_0d5c1b7a144c.png)\n![Darknet ROS 示例：检测图像](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fleggedrobotics_darknet_ros_readme_22d29d505435.png)\n\n基于 [Pascal VOC](https:\u002F\u002Fpjreddie.com\u002Fprojects\u002Fpascal-voc-dataset-mirror\u002F) 2012 数据集，YOLO 可以检测 20 种 Pascal 对象类别：\n\n- 人物\n- 鸟类、猫、牛、狗、马、羊\n- 飞机、自行车、船、公共汽车、汽车、摩托车、火车\n- 瓶子、椅子、餐桌、盆栽植物、沙发、电视\u002F显示器\n\n基于 [COCO](http:\u002F\u002Fcocodataset.org\u002F#home) 数据集，YOLO 可以检测 80 种 COCO 对象类别：\n\n- 人物\n- 自行车、汽车、摩托车、飞机、公共汽车、火车、卡车、船\n- 交通信号灯、消防栓、停车标志、停车计时器、长椅\n- 猫、狗、马、羊、牛、大象、熊、斑马、长颈鹿\n- 背包、雨伞、手提包、领带、行李箱、飞盘、滑雪板、单板滑雪、运动球、风筝、棒球棒、棒球手套、滑板、冲浪板、网球拍\n- 瓶子、葡萄酒杯、杯子、叉子、刀子、勺子、碗\n- 香蕉、苹果、三明治、橙子、西兰花、胡萝卜、热狗、披萨、甜甜圈、蛋糕\n- 椅子、沙发、盆栽植物、床、餐桌、马桶、电视显示器、笔记本电脑、鼠标、遥控器、键盘、手机、微波炉、烤箱、烤面包机、水槽、冰箱、书本、钟表、花瓶、剪刀、泰迪熊、吹风机、牙刷\n\n## 引用\n\n本软件所采用的 YOLO 方法已在论文《You Only Look Once: 统一的实时目标检测》（https:\u002F\u002Farxiv.org\u002Fabs\u002F1506.02640）中得到详细阐述。\n\n如果您正在将 YOLO V3 用于 ROS，请在您的出版物中添加以下引用：\n\nM. Bjelonic  \n**“YOLO ROS：面向 ROS 的实时目标检测”**  \n网址：https:\u002F\u002Fgithub.com\u002Fleggedrobotics\u002Fdarknet_ros，2018年。\n\n    @misc{bjelonicYolo2018,\n      author = {Marko Bjelonic},\n      title = {{YOLO ROS}: 实时目标检测 for {ROS}},\n      howpublished = {\\url{https:\u002F\u002Fgithub.com\u002Fleggedrobotics\u002Fdarknet_ros}},\n      year = {2016--2018},\n    }\n\n## 安装\n\n### 依赖项\n\n本软件基于机器人操作系统（ROS），您需要先完成 ROS 的安装步骤（详见 [ros.org](http:\u002F\u002Fwiki.ros.org)）。此外，YOLO for ROS 还依赖于以下软件：\n\n- [OpenCV](http:\u002F\u002Fopencv.org\u002F)（计算机视觉库）\n- [Boost](http:\u002F\u002Fwww.boost.org\u002F)（C++ 库）\n\n### 构建\n\n[![构建状态](https:\u002F\u002Fci.leggedrobotics.com\u002FbuildStatus\u002Ficon?job=github_leggedrobotics\u002Fdarknet_ros\u002Fmaster)](https:\u002F\u002Fci.leggedrobotics.com\u002Fjob\u002Fgithub_leggedrobotics\u002Fjob\u002Fdarknet_ros\u002Fjob\u002Fmaster\u002F)\n\n要安装 darknet_ros，请通过 SSH 克隆此仓库的最新版本（参见 [如何设置 SSH 密钥](https:\u002F\u002Fconfluence.atlassian.com\u002Fbitbucket\u002Fset-up-an-ssh-key-728138079.html)），将其克隆至您的 catkin 工作区，并使用 ROS 编译该软件包。\n\n    cd catkin_workspace\u002Fsrc\n    git clone --recursive git@github.com:leggedrobotics\u002Fdarknet_ros.git\n    cd ..\u002F\n\n为获得最佳性能，请确保以 *Release* 模式进行编译。您可以通过设置以下参数来指定编译类型：\n\n    catkin_make -DCMAKE_BUILD_TYPE=Release\n\n或使用 [Catkin 命令行工具](http:\u002F\u002Fcatkin-tools.readthedocs.io\u002Fen\u002Flatest\u002Findex.html#)\n\n    catkin build darknet_ros -DCMAKE_BUILD_TYPE=Release\n\n在 CPU 上运行 Darknet 的速度较快（约 1.5 秒，适用于 Intel Core i7-6700HQ CPU，主频 2.60GHz，配备 8 核心），但在 GPU 上的速度则快了大约 500 倍！您需要配备 NVIDIA GPU，并且必须安装 CUDA。CMakeLists.txt 文件会自动检测您是否已安装 CUDA。CUDA 是由 NVIDIA 开发的一种并行计算平台及应用程序编程接口（API）模型。若您的系统未安装 CUDA，编译过程将切换至 YOLO 的 CPU 版本。若您使用 CUDA 进行编译，可能会遇到如下构建错误：\n\n    nvcc fatal : 不支持的 GPU 架构 'compute_61'。\n\n这意味着您需要检查 GPU 的计算能力（版本）。您可以在 CUDA 官方网站上找到受支持的 GPU 列表：[CUDA - 维基百科](https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FCUDA#Supported_GPUs)。只需查找您 GPU 的计算能力，并将其添加到 darknet_ros\u002FCMakeLists.txt 中。只需添加类似如下的行：\n\n    -O3 -gencode arch=compute_62,code=sm_62\n\n### 下载权重\n\nyolo-voc.weights 和 tiny-yolo-voc.weights 会在 CMakeLists.txt 文件中自动下载。如果您需要重新下载这些权重，请进入权重文件夹，从 COCO 数据集中下载这两份预训练权重：\n\n    cd catkin_workspace\u002Fsrc\u002Fdarknet_ros\u002Fdarknet_ros\u002Fyolo_network_config\u002Fweights\u002F\n    wget http:\u002F\u002Fpjreddie.com\u002Fmedia\u002Ffiles\u002Fyolov2.weights\n    wget http:\u002F\u002Fpjreddie.com\u002Fmedia\u002Ffiles\u002Fyolov2-tiny.weights\n\n而 VOC 数据集的权重可在以下地址找到：\n\n    wget http:\u002F\u002Fpjreddie.com\u002Fmedia\u002Ffiles\u002Fyolov2-voc.weights\n    wget http:\u002F\u002Fpjreddie.com\u002Fmedia\u002Ffiles\u002Fyolov2-tiny-voc.weights\n\nYOLO v3 的预训练权重则可在此处找到：\n\n    wget http:\u002F\u002Fpjreddie.com\u002Fmedia\u002Ffiles\u002Fyolov3-tiny.weights\n    wget http:\u002F\u002Fpjreddie.com\u002Fmedia\u002Ffiles\u002Fyolov3.weights\n\n此外，还有来自不同数据集的更多预训练权重，详情请参见 [这里](https:\u002F\u002Fpjreddie.com\u002Fdarknet\u002Fyolo\u002F)。\n\n### 使用自定义检测对象\n\n要使用自定义检测对象，您需要在以下目录中提供自己的权重和配置文件：\n\n    catkin_workspace\u002Fsrc\u002Fdarknet_ros\u002Fdarknet_ros\u002Fyolo_network_config\u002Fweights\u002F\n    catkin_workspace\u002Fsrc\u002Fdarknet_ros\u002Fdarknet_ros\u002Fyolo_network_config\u002Fcfg\u002F\n\n此外，您还需为 ROS 创建一个配置文件，用于定义检测对象的名称。您需要将配置文件放置在以下路径中：\n\n    catkin_workspace\u002Fsrc\u002Fdarknet_ros\u002Fdarknet_ros\u002Fconfig\u002F\n\n然后，在启动文件中，您需要在行中指向新的配置文件：\n\n    \u003Crosparam command=\"load\" ns=\"darknet_ros\" file=\"$(find darknet_ros)\u002Fconfig\u002Fyour_config_file.yaml\"\u002F>\n\n### 单元测试\n\n使用 [Catkin 命令行工具](http:\u002F\u002Fcatkin-tools.readthedocs.io\u002Fen\u002Flatest\u002Findex.html#) 运行单元测试。\n\n    catkin build darknet_ros --no-deps --verbose --catkin-make-args run_tests\n\n您将会看到上方的图像弹出。\n\n## 基本用法\n\n要让 YOLO ROS：面向 ROS 的实时目标检测程序与您的机器人协同运行，您需要对若干参数进行调整。最简单的方法是复制并修改 `darknet_ros` 包中所需的全部参数文件。具体而言，这些参数文件包括 `config` 目录下的所有参数文件，以及 `launch` 文件夹中的启动文件。\n\n## 节点\n\n### 节点：darknet_ros\n\n这是 YOLO ROS：面向 ROS 的实时目标检测主节点。它利用摄像头采集的数据，对帧中的预训练目标进行检测。\n\n### 与 ROS 相关的参数\n\n您可以在 `darknet_ros\u002Fconfig\u002Fros.yaml` 中更改发布者、订阅者和动作的名称及其他参数。\n\n#### 订阅主题\n\n* **`\u002Fcamera_reading`** ([sensor_msgs\u002FImage])\n\n    摄像头采集的数据。\n\n#### 发布主题\n\n* **`object_detector`** ([std_msgs::Int8])\n\n    发布检测到的目标数量。\n\n* **`bounding_boxes`** ([darknet_ros_msgs::BoundingBoxes])\n\n    发布一个边界框数组，其中包含边界框在像素坐标系中的位置与尺寸信息。\n\n* **`detection_image`** ([sensor_msgs::Image])\n\n    发布包含边界框的检测图像。\n\n#### 动作\n\n* **`camera_reading`** ([sensor_msgs::Image])\n\n    发送一个带有图像的动作，其结果是一个边界框数组。\n\n### 检测相关参数\n\n您可以通过添加一个新的配置文件来调整与检测相关的参数，该文件的结构与 `darknet_ros\u002Fconfig\u002Fyolo.yaml` 类似。\n\n#### 订阅主题\n\n* **`image_view\u002Fenable_opencv`** (布尔值)\n\n    开启或关闭对检测图像（包括边界框）的 OpenCV 视图。\n\n* **`image_view\u002Fwait_key_delay`** (整数)\n\n    设置 OpenCV 窗口的等待延迟时间，以毫秒为单位。\n\n* **`yolo_model\u002Fconfig_file\u002Fname`** (字符串)\n\n    指定用于检测的网络配置文件的名称。代码会在 `darknet_ros\u002Fyolo_network_config\u002Fcfg\u002F` 目录下查找该名称。\n\n* **`yolo_model\u002Fweight_file\u002Fname`** (字符串)\n\n    指定用于检测的网络权重文件的名称。代码会在 `darknet_ros\u002Fyolo_network_config\u002Fweights\u002F` 目录下查找该名称。\n\n* **`yolo_model\u002Fthreshold\u002Fvalue`** (浮点数)\n\n    检测算法的阈值，范围为 0 到 1。\n\n* **`yolo_model\u002Fdetection_classes\u002Fnames`** (字符串数组)\n\n    指定用于检测的网络的检测名称，这些名称存储在 `darknet_ros\u002Fyolo_network_config\u002F` 目录下的配置文件和权重文件中。","# darknet_ros 快速上手指南\n\n## 环境准备\n\n| 项目 | 要求 |\n|---|---|\n| 操作系统 | Ubuntu 20.04（推荐） |\n| ROS 版本 | ROS Noetic（已验证） |\n| GPU（可选） | NVIDIA GPU + CUDA |\n| 依赖 | ROS、OpenCV、boost |\n\n> 若使用 ROS Melodic \u002F ROS2 Foxy，请切换到对应分支。\n\n## 安装步骤\n\n1. **克隆源码**  \n   ```bash\n   cd ~\u002Fcatkin_ws\u002Fsrc\n   git clone --recursive git@github.com:leggedrobotics\u002Fdarknet_ros.git\n   cd ..\n   ```\n\n2. **编译（Release 模式）**  \n   ```bash\n   catkin_make -DCMAKE_BUILD_TYPE=Release\n   # 或使用 catkin-tools\n   # catkin build darknet_ros -DCMAKE_BUILD_TYPE=Release\n   ```\n\n3. **下载权重（自动完成）**  \n   如需手动下载：\n   ```bash\n   cd ~\u002Fcatkin_ws\u002Fsrc\u002Fdarknet_ros\u002Fdarknet_ros\u002Fyolo_network_config\u002Fweights\u002F\n   wget http:\u002F\u002Fpjreddie.com\u002Fmedia\u002Ffiles\u002Fyolov3.weights\n   wget http:\u002F\u002Fpjreddie.com\u002Fmedia\u002Ffiles\u002Fyolov3-tiny.weights\n   ```\n\n## 基本使用\n\n1. **启动节点**  \n   ```bash\n   roslaunch darknet_ros darknet_ros.launch\n   ```\n\n2. **订阅\u002F发布话题**  \n   - 订阅：`\u002Fcamera_reading`（sensor_msgs\u002FImage）  \n   - 发布：\n     - `\u002Fdarknet_ros\u002Fbounding_boxes`（检测框）  \n     - `\u002Fdarknet_ros\u002Fdetection_image`（带框图像）\n\n3. **快速验证**  \n   使用 bag 包测试：\n   ```bash\n   rosbag play test.bag  # 包含 \u002Fcamera_reading\n   rviz  # 订阅 \u002Fdarknet_ros\u002Fdetection_image 查看结果\n   ```\n\n4. **自定义检测类别**  \n   - 将权重 `.weights` 和配置 `.cfg` 放入  \n     `yolo_network_config\u002Fweights\u002F` 与 `cfg\u002F`  \n   - 新建 `config\u002Fmy_objects.yaml` 并修改 launch 文件指向即可。","某高校机器人实验室正在开发一台校园巡检机器人，需要在室外道路和教学楼内实时识别行人、自行车、消防栓等关键目标，以便自动避障并上报异常。\n\n### 没有 darknet_ros 时\n- 工程师先用 PyTorch 训练好 YOLOv3，再手写 ROS 节点把 `.weights` 转成 `cv::Mat`，结果 GPU 显存频繁爆掉，帧率掉到 5 fps。  \n- 为了把检测结果发出去，又写了一个自定义消息类型，结果和导航栈的 `costmap_2d` 接口对不上，每改一次消息格式就要重新编译 20 分钟。  \n- 在室外测试时，阳光直射导致图像过曝，检测框漂移严重；团队只能手动调 OpenCV 的亮度\u002F对比度，调完还得重新标定相机内参。  \n- 机器人 CPU 占用飙到 90%，风扇狂转，续航从 3 小时降到 1 小时，最后被迫把图像分辨率从 640×480 缩到 320×240，检测距离缩短一半。  \n\n### 使用 darknet_ros 后\n- 一行 `roslaunch darknet_ros yolo_v3.launch` 就启动 GPU 加速，默认 30 fps，显存占用稳定在 2 GB，直接发布 `\u002Fdarknet_ros\u002Fbounding_boxes`，无需再写任何 C++。  \n- 检测结果以标准 `vision_msgs\u002FDetection2DArray` 输出，导航栈直接订阅即可生成避障代价地图，节省 3 天接口联调时间。  \n- 内置的图像预处理节点自动做白平衡和直方图均衡，过曝场景下行人检测召回率从 72% 提升到 91%，不用再手动调参。  \n- CPU 占用降到 25%，续航恢复到 2.5 小时；通过 `param\u002Fyolo.yaml` 把输入分辨率改回 640×480，检测距离恢复到 20 m，夜间还能识别 15 m 外的消防栓。  \n\ndarknet_ros 让实验室在一周内就把“看得见”的巡检机器人原型跑通，把工程师从底层适配中解放出来，专注做上层业务逻辑。","https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fleggedrobotics_darknet_ros_0d5c1b7a.png","leggedrobotics","Robotic Systems Lab - Legged Robotics at ETH Zürich","https:\u002F\u002Foss.gittoolsai.com\u002Favatars\u002Fleggedrobotics_1f89d414.png","The Robotic Systems Lab investigates the development of machines and their intelligence to operate in rough and challenging environments. ",null,"https:\u002F\u002Frsl.ethz.ch\u002F","https:\u002F\u002Fgithub.com\u002Fleggedrobotics",[83,87],{"name":84,"color":85,"percentage":86},"C++","#f34b7d",78.9,{"name":88,"color":89,"percentage":90},"CMake","#DA3434",21.1,2425,1207,"2026-04-04T14:03:30","BSD-3-Clause",4,"Linux","可选；若使用 GPU，需 NVIDIA GPU 并安装 CUDA（具体版本未说明，需根据 GPU 计算能力调整 CMakeLists.txt）","未说明",{"notes":100,"python":98,"dependencies":101},"已在 Ubuntu 20.04 + ROS Noetic 测试通过；也提供 ROS Melodic、ROS Foxy 和 ROS2 分支。若使用 GPU，需安装 CUDA，否则自动回退到 CPU 版本。首次编译时会自动下载 yolo-voc.weights 和 tiny-yolo-voc.weights。",[102,103,104],"ROS Noetic","OpenCV","boost",[13,14],[107,108,109,110,111,112,113,114],"object-detection","human-detection","deep-learning","yolo","darknet","ros","computer-vision","darknet-ros","2026-03-27T02:49:30.150509","2026-04-06T07:13:18.902651",[118,123,128,133,138,143,148,153],{"id":119,"question_zh":120,"answer_zh":121,"source_url":122},6247,"编译时报错 “OpenCV 4.x+ requires enabled C++11 support” 如何解决？","官方主分支对 OpenCV4 支持滞后。可切换到社区维护的 opencv4 分支，例如 https:\u002F\u002Fgithub.com\u002Fkunaltyagi\u002Fdarknet_ros 的 opencv4 分支（需带 submodules 克隆），或参考 PR #257、#230 的补丁。","https:\u002F\u002Fgithub.com\u002Fleggedrobotics\u002Fdarknet_ros\u002Fissues\u002F200",{"id":124,"question_zh":125,"answer_zh":126,"source_url":127},6244,"在 Jetson TX2 上运行 darknet_ros 的 YOLO 和 tiny-YOLO 能达到多少帧率？","在 TX2 上使用 devel\u002Fthreads 分支，tiny-yolo.cfg 输入 854×480 分辨率时，实测可达 17 FPS（416×416 网络输入）。","https:\u002F\u002Fgithub.com\u002Fleggedrobotics\u002Fdarknet_ros\u002Fissues\u002F51",{"id":129,"question_zh":130,"answer_zh":131,"source_url":132},6245,"为什么我已经用 Release 模式编译，darknet_ros 的帧率仍然比原生 darknet 低？","帧率差异主要来自输入分辨率：1080p 的像素数是 640×480 的 6.75 倍，若假设运行时间与像素数成线性关系，帧率会相应下降。请确认输入分辨率，并用 nvidia-smi 与 top 查看 GPU\u002FCPU 占用，排除瓶颈。","https:\u002F\u002Fgithub.com\u002Fleggedrobotics\u002Fdarknet_ros\u002Fissues\u002F25",{"id":134,"question_zh":135,"answer_zh":136,"source_url":137},6246,"如何利用 darknet_ros 与 Kinect 深度图估计目标的 3D 距离？","YOLO 只给出 2D 框 (u,v)。可订阅 Kinect 的 PointCloud2 话题，将 2D 框中心 (u,v) 映射到点云，直接读取对应点的 (x,y,z) 即可得到相机坐标系下的 3D 位置。","https:\u002F\u002Fgithub.com\u002Fleggedrobotics\u002Fdarknet_ros\u002Fissues\u002F103",{"id":139,"question_zh":140,"answer_zh":141,"source_url":142},6248,"运行 darknet_ros 时出现内存泄漏，30 秒内占满 8 GB RAM 怎么办？","该问题已在后续 commit 中修复。请更新到最新代码（issue #311 已关闭并合并补丁），重新编译即可解决内存泄漏导致的节点崩溃。","https:\u002F\u002Fgithub.com\u002Fleggedrobotics\u002Fdarknet_ros\u002Fissues\u002F311",{"id":144,"question_zh":145,"answer_zh":146,"source_url":147},6249,"darknet_ros 会支持 YOLO v4 吗？","维护者已确认会支持 YOLO v4。目前社区已有可编译通过的 fork（如 https:\u002F\u002Fgithub.com\u002Ftom13133\u002Fdarknet_ros），但在 Jetson Xavier NX 上 YOLOv4 仅 2–3 FPS，比 YOLOv3 的 4–5 FPS 慢，可按需选择。","https:\u002F\u002Fgithub.com\u002Fleggedrobotics\u002Fdarknet_ros\u002Fissues\u002F243",{"id":149,"question_zh":150,"answer_zh":151,"source_url":152},6250,"用自定义数据训练的 .weights 在 darknet 能检测，在 darknet_ros 却检测不到怎么办？","确认已正确放置 .cfg、.weights、.yaml 并修改 launch 文件。若仍无检测，可尝试：1) 检查类别名、anchor 与 .yaml 是否一致；2) 降低 confidence threshold；3) 若仍失败，可暂时换用其他 ROS-YOLO 实现（issue 中用户提到 5 个月前曾成功换用替代方案）。","https:\u002F\u002Fgithub.com\u002Fleggedrobotics\u002Fdarknet_ros\u002Fissues\u002F307",{"id":154,"question_zh":155,"answer_zh":156,"source_url":157},6251,"如何把 ZED 相机接入 darknet_ros？","先确保 ZED SDK 与官方 ROS wrapper 已正常工作。然后在 darknet_ros 的 launch 文件里把 image 话题 remap 到 \u002Fzed\u002Fzed_node\u002Frgb\u002Fimage_rect_color（或对应左右目话题），即可直接订阅 ZED 图像进行目标检测。若出现红蓝通道互换，可在 ZED wrapper 中设置 enable_rgb_swap=true 解决。","https:\u002F\u002Fgithub.com\u002Fleggedrobotics\u002Fdarknet_ros\u002Fissues\u002F105",[159,163],{"id":160,"version":161,"summary_zh":79,"released_at":162},105803,"1.1.5","2021-04-08T20:10:43",{"id":164,"version":165,"summary_zh":79,"released_at":166},105804,"1.1.4","2020-04-14T14:56:55"]