[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"similar-obss--sahi":3,"tool-obss--sahi":61},[4,18,26,36,44,53],{"id":5,"name":6,"github_repo":7,"description_zh":8,"stars":9,"difficulty_score":10,"last_commit_at":11,"category_tags":12,"status":17},4358,"openclaw","openclaw\u002Fopenclaw","OpenClaw 是一款专为个人打造的本地化 AI 助手，旨在让你在自己的设备上拥有完全可控的智能伙伴。它打破了传统 AI 助手局限于特定网页或应用的束缚，能够直接接入你日常使用的各类通讯渠道，包括微信、WhatsApp、Telegram、Discord、iMessage 等数十种平台。无论你在哪个聊天软件中发送消息，OpenClaw 都能即时响应，甚至支持在 macOS、iOS 和 Android 设备上进行语音交互，并提供实时的画布渲染功能供你操控。\n\n这款工具主要解决了用户对数据隐私、响应速度以及“始终在线”体验的需求。通过将 AI 部署在本地，用户无需依赖云端服务即可享受快速、私密的智能辅助，真正实现了“你的数据，你做主”。其独特的技术亮点在于强大的网关架构，将控制平面与核心助手分离，确保跨平台通信的流畅性与扩展性。\n\nOpenClaw 非常适合希望构建个性化工作流的技术爱好者、开发者，以及注重隐私保护且不愿被单一生态绑定的普通用户。只要具备基础的终端操作能力（支持 macOS、Linux 及 Windows WSL2），即可通过简单的命令行引导完成部署。如果你渴望拥有一个懂你",349277,3,"2026-04-06T06:32:30",[13,14,15,16],"Agent","开发框架","图像","数据工具","ready",{"id":19,"name":20,"github_repo":21,"description_zh":22,"stars":23,"difficulty_score":10,"last_commit_at":24,"category_tags":25,"status":17},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,"2026-04-05T11:01:52",[14,15,13],{"id":27,"name":28,"github_repo":29,"description_zh":30,"stars":31,"difficulty_score":32,"last_commit_at":33,"category_tags":34,"status":17},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",160015,2,"2026-04-18T11:30:52",[14,13,35],"语言模型",{"id":37,"name":38,"github_repo":39,"description_zh":40,"stars":41,"difficulty_score":32,"last_commit_at":42,"category_tags":43,"status":17},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。",109154,"2026-04-18T11:18:24",[14,15,13],{"id":45,"name":46,"github_repo":47,"description_zh":48,"stars":49,"difficulty_score":32,"last_commit_at":50,"category_tags":51,"status":17},6121,"gemini-cli","google-gemini\u002Fgemini-cli","gemini-cli 是一款由谷歌推出的开源 AI 命令行工具，它将强大的 Gemini 大模型能力直接集成到用户的终端环境中。对于习惯在命令行工作的开发者而言，它提供了一条从输入提示词到获取模型响应的最短路径，无需切换窗口即可享受智能辅助。\n\n这款工具主要解决了开发过程中频繁上下文切换的痛点，让用户能在熟悉的终端界面内直接完成代码理解、生成、调试以及自动化运维任务。无论是查询大型代码库、根据草图生成应用，还是执行复杂的 Git 操作，gemini-cli 都能通过自然语言指令高效处理。\n\n它特别适合广大软件工程师、DevOps 人员及技术研究人员使用。其核心亮点包括支持高达 100 万 token 的超长上下文窗口，具备出色的逻辑推理能力；内置 Google 搜索、文件操作及 Shell 命令执行等实用工具；更独特的是，它支持 MCP（模型上下文协议），允许用户灵活扩展自定义集成，连接如图像生成等外部能力。此外，个人谷歌账号即可享受免费的额度支持，且项目基于 Apache 2.0 协议完全开源，是提升终端工作效率的理想助手。",100752,"2026-04-10T01:20:03",[52,13,15,14],"插件",{"id":54,"name":55,"github_repo":56,"description_zh":57,"stars":58,"difficulty_score":32,"last_commit_at":59,"category_tags":60,"status":17},4721,"markitdown","microsoft\u002Fmarkitdown","MarkItDown 是一款由微软 AutoGen 团队打造的轻量级 Python 工具，专为将各类文件高效转换为 Markdown 格式而设计。它支持 PDF、Word、Excel、PPT、图片（含 OCR）、音频（含语音转录）、HTML 乃至 YouTube 链接等多种格式的解析，能够精准提取文档中的标题、列表、表格和链接等关键结构信息。\n\n在人工智能应用日益普及的今天，大语言模型（LLM）虽擅长处理文本，却难以直接读取复杂的二进制办公文档。MarkItDown 恰好解决了这一痛点，它将非结构化或半结构化的文件转化为模型“原生理解”且 Token 效率极高的 Markdown 格式，成为连接本地文件与 AI 分析 pipeline 的理想桥梁。此外，它还提供了 MCP（模型上下文协议）服务器，可无缝集成到 Claude Desktop 等 LLM 应用中。\n\n这款工具特别适合开发者、数据科学家及 AI 研究人员使用，尤其是那些需要构建文档检索增强生成（RAG）系统、进行批量文本分析或希望让 AI 助手直接“阅读”本地文件的用户。虽然生成的内容也具备一定可读性，但其核心优势在于为机器",93400,"2026-04-06T19:52:38",[52,14],{"id":62,"github_repo":63,"name":64,"description_en":65,"description_zh":66,"ai_summary_zh":66,"readme_en":67,"readme_zh":68,"quickstart_zh":69,"use_case_zh":70,"hero_image_url":71,"owner_login":72,"owner_name":73,"owner_avatar_url":74,"owner_bio":75,"owner_company":76,"owner_location":76,"owner_email":77,"owner_twitter":78,"owner_website":79,"owner_url":80,"languages":81,"stars":90,"forks":91,"last_commit_at":92,"license":93,"difficulty_score":32,"env_os":94,"env_gpu":95,"env_ram":94,"env_deps":96,"category_tags":109,"github_topics":110,"view_count":32,"oss_zip_url":76,"oss_zip_packed_at":76,"status":17,"created_at":130,"updated_at":131,"faqs":132,"releases":133},9293,"obss\u002Fsahi","sahi","Framework agnostic sliced\u002Ftiled inference + interactive ui + error analysis plots","SAHI（Slicing Aided Hyper Inference）是一款轻量级的视觉算法库，专为大规模目标检测与实例分割任务设计。在现实应用中，当处理高分辨率大图（如卫星遥感、医疗影像或工业质检图片）时，传统模型往往难以精准识别其中的微小目标，容易出现漏检或定位不准的情况。SAHI 通过独特的“切片推理”技术巧妙解决了这一难题：它将大图像智能切割成多个重叠的小片段分别进行检测，随后再将结果无缝合并，从而显著提升对小目标的识别精度。\n\n该工具具有极强的兼容性，不绑定特定的深度学习框架，能够轻松集成 YOLO、MMDetection 等主流检测模型。除了核心推理功能，SAHI 还配备了交互式可视化界面和误差分析图表，帮助开发者直观地调试模型表现并优化效果。\n\nSAHI 非常适合计算机视觉开发者、算法研究人员以及需要处理海量高清图像数据的工程师使用。无论是进行学术研究还是落地工业应用，只需几行代码即可调用其简洁的 API，高效完成从数据预处理到结果分析的全流程，是提升大图小目标检测能力的得力助手。","\u003Cdiv align=\"center\">\n\u003Ch1>\n SAHI: Slicing Aided Hyper Inference\n\u003C\u002Fh1>\n\n\u003Ch4>\n A lightweight vision library for performing large scale object detection & instance segmentation\n\u003C\u002Fh4>\n\n\u003Ch4>\n \u003Cimg width=\"700\" alt=\"teaser\" src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fobss_sahi_readme_97f7e16a88ae.png\">\n\u003C\u002Fh4>\n\n\u003C!-- Downloads & Version -->\n\u003Cdiv>\n \u003Ca href=\"https:\u002F\u002Fpepy.tech\u002Fproject\u002Fsahi\">\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fobss_sahi_readme_fae08394318e.png\" alt=\"Total Downloads\">\u003C\u002Fa>\n \u003Ca href=\"https:\u002F\u002Fpepy.tech\u002Fproject\u002Fsahi\">\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fobss_sahi_readme_fae08394318e.png\u002Fmonth\" alt=\"Monthly Downloads\">\u003C\u002Fa>\n \u003Ca href=\"https:\u002F\u002Fbadge.fury.io\u002Fpy\u002Fsahi\">\u003Cimg src=\"https:\u002F\u002Fbadge.fury.io\u002Fpy\u002Fsahi.svg\" alt=\"PyPI Version\">\u003C\u002Fa>\n \u003Ca href=\"https:\u002F\u002Fanaconda.org\u002Fconda-forge\u002Fsahi\">\u003Cimg src=\"https:\u002F\u002Fanaconda.org\u002Fconda-forge\u002Fsahi\u002Fbadges\u002Fversion.svg\" alt=\"Conda Version\">\u003C\u002Fa>\n \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fblob\u002Fmain\u002FLICENSE.md\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fpypi\u002Fl\u002Fsahi\" alt=\"License\">\u003C\u002Fa>\n\u003C\u002Fdiv>\n\n\u003C!-- CI & Quality -->\n\u003Cdiv>\n \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Factions\u002Fworkflows\u002Fci.yml\">\u003Cimg src=\"https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Factions\u002Fworkflows\u002Fci.yml\u002Fbadge.svg\" alt=\"CI\">\u003C\u002Fa>\n \u003Ca href=\"https:\u002F\u002Fsecurity.snyk.io\u002Fpackage\u002Fpip\u002Fsahi\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FSnyk_security-monitored-8A2BE2\" alt=\"Known Vulnerabilities\">\u003C\u002Fa>\n \u003Ca href=\"https:\u002F\u002Fwww.codefactor.io\u002Frepository\u002Fgithub\u002Fonuralpszr\u002Fsahi\">\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fobss_sahi_readme_9ee0cb95ac54.png\" alt=\"CodeFactor\">\u003C\u002Fa>\n \u003Ca href=\"https:\u002F\u002Fieeexplore.ieee.org\u002Fdocument\u002F9897990\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FDOI-10.1109%2FICIP46576.2022.9897990-orange.svg\" alt=\"DOI\">\u003C\u002Fa>\n\u003C\u002Fdiv>\n\n\u003C!-- AI & Docs -->\n\u003Cdiv>\n \u003Ca href=\"https:\u002F\u002Fcontext7.com\u002Fobss\u002Fsahi\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FContext7%20MCP-Indexed-blue\" alt=\"Context7 MCP\">\u003C\u002Fa>\n \u003Ca href=\"https:\u002F\u002Fcontext7.com\u002Fobss\u002Fsahi\u002Fllms.txt\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fllms.txt-✓-brightgreen\" alt=\"llms.txt\">\u003C\u002Fa>\n \u003Ca href=\"https:\u002F\u002Fdeepwiki.com\u002Fobss\u002Fsahi\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FDeepWiki-obss%2Fsahi-blue.svg?logo=data:image\u002Fpng;base64,iVBORw0KGgoAAAANSUhEUgAAACwAAAAyCAYAAAAnWDnqAAAAAXNSR0IArs4c6QAAA05JREFUaEPtmUtyEzEQhtWTQyQLHNak2AB7ZnyXZMEjXMGeK\u002FAIi+QuHrMnbChYY7MIh8g01fJoopFb0uhhEqqcbWTp06\u002Fuv1saEDv4O3n3dV60RfP947Mm9\u002FSQc0ICFQgzfc4CYZoTPAswgSJCCUJUnAAoRHOAUOcATwbmVLWdGoH\u002F\u002FPB8mnKqScAhsD0kYP3j\u002FYt5LPQe2KvcXmGvRHcDnpxfL2zOYJ1mFwrryWTz0advv1Ut4CJgf5uhDuDj5eUcAUoahrdY\u002F56ebRWeraTjMt\u002F00Sh3UDtjgHtQNHwcRGOC98BJEAEymycmYcWwOprTgcB6VZ5JK5TAJ+fXGLBm3FDAmn6oPPjR4rKCAoJCal2eAiQp2x0vxTPB3ALO2CRkwmDy5WohzBDwSEFKRwPbknEggCPB\u002FimwrycgxX2NzoMCHhPkDwqYMr9tRcP5qNrMZHkVnOjRMWwLCcr8ohBVb1OMjxLwGCvjTikrsBOiA6fNyCrm8V1rP93iVPpwaE+gO0SsWmPiXB+jikdf6SizrT5qKasx5j8ABbHpFTx+vFXp9EnYQmLx02h1QTTrl6eDqxLnGjporxl3NL3agEvXdT0WmEost648sQOYAeJS9Q7bfUVoMGnjo4AZdUMQku50McDcMWcBPvr0SzbTAFDfvJqwLzgxwATnCgnp4wDl6Aa+Ax283gghmj+vj7feE2KBBRMW3FzOpLOADl0Isb5587h\u002FU4gGvkt5v60Z1VLG8BhYjbzRwyQZemwAd6cCR5\u002FXFWLYZRIMpX39AR0tjaGGiGzLVyhse5C9RKC6ai42ppWPKiBagOvaYk8lO7DajerabOZP46Lby5wKjw1HCRx7p9sVMOWGzb\u002FvA1hwiWc6jm3MvQDTogQkiqIhJV0nBQBTU+3okKCFDy9WwferkHjtxib7t3xIUQtHxnIwtx4mpg26\u002FHfwVNVDb4oI9RHmx5WGelRVlrtiw43zboCLaxv46AZeB3IlTkwouebTr1y2NjSpHz68WNFjHvupy3q8TFn3Hos2IAk4Ju5dCo8B3wP7VPr\u002FFGaKiG+T+v+TQqIrOqMTL1VdWV1DdmcbO8KXBz6esmYWYKPwDL5b5FA1a0hwapHiom0r\u002FcKaoqr+27\u002FXcrS5UwSMbQAAAABJRU5ErkJggg==\" alt=\"DeepWiki\">\u003C\u002Fa>\n \u003Ca href=\"https:\u002F\u002Fhuggingface.co\u002Fspaces\u002Ffcakyon\u002Fsahi-yolox\">\u003Cimg src=\"https:\u002F\u002Fraw.githubusercontent.com\u002Fobss\u002Fsahi\u002Fmain\u002Fresources\u002Fhf_spaces_badge.svg\" alt=\"HuggingFace Spaces\">\u003C\u002Fa>\n\u003C\u002Fdiv>\n\n\u003C\u002Fdiv>\n\n## \u003Cdiv align=\"center\">Overview\u003C\u002Fdiv>\n\nSAHI helps developers overcome real-world challenges in object detection by\nenabling **sliced inference** for detecting small objects in large images. It\nsupports various popular detection models and provides easy-to-use APIs.\n\n\u003Cdiv align=\"center\">\n\n🌐 [English](README.md) | 🇨🇳 [简体中文](docs\u002Fzh\u002FREADME.md)\n\n\u003C\u002Fdiv>\n\n| Command | Description |\n| ----------------------------------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |\n| [predict](https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fblob\u002Fmain\u002Fdocs\u002Fcli.md#predict-command-usage) | Perform sliced\u002Fstandard video\u002Fimage prediction using any [ultralytics](https:\u002F\u002Fgithub.com\u002Fultralytics\u002Fultralytics) \u002F [mmdet](https:\u002F\u002Fgithub.com\u002Fopen-mmlab\u002Fmmdetection) \u002F [huggingface](https:\u002F\u002Fhuggingface.co\u002Fmodels?pipeline_tag=object-detection&sort=downloads) \u002F [torchvision](https:\u002F\u002Fpytorch.org\u002Fvision\u002Fstable\u002Fmodels.html#object-detection) model — see [CLI guide](docs\u002Fcli.md#predict-command-usage) |\n| [predict-fiftyone](https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fblob\u002Fmain\u002Fdocs\u002Fcli.md#predict-fiftyone-command-usage) | Perform sliced\u002Fstandard prediction using any supported model and explore results in [fiftyone app](https:\u002F\u002Fgithub.com\u002Fvoxel51\u002Ffiftyone) — [learn more](docs\u002Ffiftyone.md) |\n| [coco slice](https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fblob\u002Fmain\u002Fdocs\u002Fcli.md#coco-slice-command-usage) | Automatically slice COCO annotation and image files — see [slicing utilities](docs\u002Fslicing.md) |\n| [coco fiftyone](https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fblob\u002Fmain\u002Fdocs\u002Fcli.md#coco-fiftyone-command-usage) | Explore multiple prediction results on your COCO dataset with [fiftyone ui](https:\u002F\u002Fgithub.com\u002Fvoxel51\u002Ffiftyone) ordered by number of misdetections |\n| [coco evaluate](https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fblob\u002Fmain\u002Fdocs\u002Fcli.md#coco-evaluate-command-usage) | Evaluate classwise COCO AP and AR for given predictions and ground truth — check [COCO utilities](docs\u002Fcoco.md) |\n| [coco analyse](https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fblob\u002Fmain\u002Fdocs\u002Fcli.md#coco-analyse-command-usage) | Calculate and export many error analysis plots — see the [complete guide](docs\u002FREADME.md) |\n| [coco yolo](https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fblob\u002Fmain\u002Fdocs\u002Fcli.md#coco-yolo-command-usage) | Automatically convert any COCO dataset to [ultralytics](https:\u002F\u002Fgithub.com\u002Fultralytics\u002Fultralytics) format |\n\n### Approved by the Community\n\n[📜 List of publications that cite SAHI (currently 600+)](https:\u002F\u002Fscholar.google.com\u002Fscholar?hl=en&as_sdt=2005&sciodt=0,5&cites=14065474760484865747&scipsc=&q=&scisbd=1)\n\n[🏆 List of competition winners that used SAHI](https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fdiscussions\u002F688)\n\n### Approved by AI Tools\n\nSAHI's documentation is\n[indexed in Context7 MCP](https:\u002F\u002Fcontext7.com\u002Fobss\u002Fsahi), providing AI coding\nassistants with up-to-date, version-specific code examples and API references.\nWe also provide an [llms.txt](https:\u002F\u002Fcontext7.com\u002Fobss\u002Fsahi\u002Fllms.txt) file\nfollowing the emerging standard for AI-readable documentation. To integrate SAHI\ndocs with your AI development workflow, check out the\n[Context7 MCP installation guide](https:\u002F\u002Fgithub.com\u002Fupstash\u002Fcontext7#%EF%B8%8F-installation).\n\n## \u003Cdiv align=\"center\">Installation\u003C\u002Fdiv>\n\n### Basic Installation\n\n```bash\npip install sahi\n```\n\n\u003Cdetails closed>\n\u003Csummary>\n\u003Cbig>\u003Cb>Detailed Installation (Click to open)\u003C\u002Fb>\u003C\u002Fbig>\n\u003C\u002Fsummary>\n\n- Install your desired version of pytorch and torchvision:\n\n```console\npip install torch==2.7.0 torchvision==0.22.0 --index-url https:\u002F\u002Fdownload.pytorch.org\u002Fwhl\u002Fcu126\n```\n\n(torch 2.1.2 is required for mmdet support):\n\n```console\npip install torch==2.1.2 torchvision==0.16.2 --index-url https:\u002F\u002Fdownload.pytorch.org\u002Fwhl\u002Fcu121\n```\n\n- Install your desired detection framework (ultralytics):\n\n```console\npip install ultralytics>=8.3.161\n```\n\n- Install your desired detection framework (huggingface):\n\n```console\npip install transformers>=4.49.0 timm\n```\n\n- Install your desired detection framework (yolov5):\n\n```console\npip install yolov5==7.0.14 sahi==0.11.21\n```\n\n- Install your desired detection framework (mmdet):\n\n```console\npip install mim\nmim install mmdet==3.3.0\n```\n\n- Install your desired detection framework (roboflow):\n\n```console\npip install inference>=0.51.5 rfdetr>=1.6.2\n```\n\n\u003C\u002Fdetails>\n\n## \u003Cdiv align=\"center\">Quick Start\u003C\u002Fdiv>\n\n### Learning Resources\n\n| Resource | Type |\n| --------------------------------------------------------------------------------------------------------------------------------------------------- | ---------- |\n| [Introduction to SAHI](https:\u002F\u002Fmedium.com\u002Fcodable\u002Fsahi-a-vision-library-for-performing-sliced-inference-on-large-images-small-objects-c8b086af3b80) | Blog Post |\n| [2025 Video Tutorial](https:\u002F\u002Fwww.youtube.com\u002Fwatch?v=ILqMBah5ZvI) ⭐ | Video |\n| [Official Paper](https:\u002F\u002Fieeexplore.ieee.org\u002Fdocument\u002F9897990) (ICIP 2022 oral) | Paper |\n| [Pretrained Weights & ICIP 2022 Paper Files](https:\u002F\u002Fgithub.com\u002Ffcakyon\u002Fsmall-object-detection-benchmark) | Benchmark |\n| [Visualizing and Evaluating SAHI Predictions with FiftyOne](https:\u002F\u002Fvoxel51.com\u002Fblog\u002Fhow-to-detect-small-objects\u002F) | Blog Post |\n| [Exploring SAHI – learnopencv.com](https:\u002F\u002Flearnopencv.com\u002Fslicing-aided-hyper-inference\u002F) | Article |\n| [Slicing Aided Hyper Inference Explained by Encord](https:\u002F\u002Fencord.com\u002Fblog\u002Fslicing-aided-hyper-inference-explained\u002F) | Article |\n| [Video Tutorial: SAHI for Small Object Detection](https:\u002F\u002Fwww.youtube.com\u002Fwatch?v=UuOJKxn-M8&t=270s) | Video |\n| [Satellite Object Detection](https:\u002F\u002Fblog.ml6.eu\u002Fhow-to-detect-small-objects-in-very-large-images-70234bab0f98) | Blog Post |\n| [COCO Dataset Conversion](https:\u002F\u002Fmedium.com\u002Fcodable\u002Fconvert-any-dataset-to-coco-object-detection-format-with-sahi-95349e1fe2b7) | Blog Post |\n| [Kaggle Notebook](https:\u002F\u002Fwww.kaggle.com\u002Fremekkinas\u002Fsahi-slicing-aided-hyper-inference-yv5-and-yx) | Notebook |\n| [Error Analysis Plots & Evaluation](https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fdiscussions\u002F622) ⭐ | Discussion |\n| [Interactive Result Visualization and Inspection](https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fdiscussions\u002F624) ⭐ | Discussion |\n| [Video Inference Support](https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fdiscussions\u002F626) | Discussion |\n| [Slicing Operation Notebook](demo\u002Fslicing.ipynb) | Notebook |\n| [Complete Documentation](docs\u002FREADME.md) | Docs |\n\n### Notebooks & Demos\n\n| Framework | Notebook | Demo |\n| ------------------ | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------- |\n| YOLO12 | [![Open In Colab](https:\u002F\u002Fcolab.research.google.com\u002Fassets\u002Fcolab-badge.svg)](https:\u002F\u002Fcolab.research.google.com\u002Fgithub\u002Fobss\u002Fsahi\u002Fblob\u002Fmain\u002Fdemo\u002Finference_for_ultralytics.ipynb) | — |\n| YOLO11 | [![Open In Colab](https:\u002F\u002Fcolab.research.google.com\u002Fassets\u002Fcolab-badge.svg)](https:\u002F\u002Fcolab.research.google.com\u002Fgithub\u002Fobss\u002Fsahi\u002Fblob\u002Fmain\u002Fdemo\u002Finference_for_ultralytics.ipynb) | — |\n| YOLO11-OBB | [![Open In Colab](https:\u002F\u002Fcolab.research.google.com\u002Fassets\u002Fcolab-badge.svg)](https:\u002F\u002Fcolab.research.google.com\u002Fgithub\u002Fobss\u002Fsahi\u002Fblob\u002Fmain\u002Fdemo\u002Finference_for_ultralytics.ipynb) | — |\n| Roboflow \u002F RF-DETR | [![Open In Colab](https:\u002F\u002Fcolab.research.google.com\u002Fassets\u002Fcolab-badge.svg)](https:\u002F\u002Fcolab.research.google.com\u002Fgithub\u002Fobss\u002Fsahi\u002Fblob\u002Fmain\u002Fdemo\u002Finference_for_roboflow.ipynb) | — |\n| RT-DETR v2 | [![Open In Colab](https:\u002F\u002Fcolab.research.google.com\u002Fassets\u002Fcolab-badge.svg)](https:\u002F\u002Fcolab.research.google.com\u002Fgithub\u002Fobss\u002Fsahi\u002Fblob\u002Fmain\u002Fdemo\u002Finference_for_huggingface.ipynb) | — |\n| RT-DETR | [![Open In Colab](https:\u002F\u002Fcolab.research.google.com\u002Fassets\u002Fcolab-badge.svg)](https:\u002F\u002Fcolab.research.google.com\u002Fgithub\u002Fobss\u002Fsahi\u002Fblob\u002Fmain\u002Fdemo\u002Finference_for_rtdetr.ipynb) | — |\n| HuggingFace | [![Open In Colab](https:\u002F\u002Fcolab.research.google.com\u002Fassets\u002Fcolab-badge.svg)](https:\u002F\u002Fcolab.research.google.com\u002Fgithub\u002Fobss\u002Fsahi\u002Fblob\u002Fmain\u002Fdemo\u002Finference_for_huggingface.ipynb) | — |\n| YOLOv5 | [![Open In Colab](https:\u002F\u002Fcolab.research.google.com\u002Fassets\u002Fcolab-badge.svg)](https:\u002F\u002Fcolab.research.google.com\u002Fgithub\u002Fobss\u002Fsahi\u002Fblob\u002Fmain\u002Fdemo\u002Finference_for_yolov5.ipynb) | — |\n| MMDetection | [![Open In Colab](https:\u002F\u002Fcolab.research.google.com\u002Fassets\u002Fcolab-badge.svg)](https:\u002F\u002Fcolab.research.google.com\u002Fgithub\u002Fobss\u002Fsahi\u002Fblob\u002Fmain\u002Fdemo\u002Finference_for_mmdetection.ipynb) | — |\n| TorchVision | [![Open In Colab](https:\u002F\u002Fcolab.research.google.com\u002Fassets\u002Fcolab-badge.svg)](https:\u002F\u002Fcolab.research.google.com\u002Fgithub\u002Fobss\u002Fsahi\u002Fblob\u002Fmain\u002Fdemo\u002Finference_for_torchvision.ipynb) | — |\n| YOLOX | — | [![HuggingFace Spaces](https:\u002F\u002Fraw.githubusercontent.com\u002Fobss\u002Fsahi\u002Fmain\u002Fresources\u002Fhf_spaces_badge.svg)](https:\u002F\u002Fhuggingface.co\u002Fspaces\u002Ffcakyon\u002Fsahi-yolox) |\n\n\u003Ca href=\"https:\u002F\u002Fhuggingface.co\u002Fspaces\u002Ffcakyon\u002Fsahi-yolox\">\u003Cimg width=\"600\" src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fobss_sahi_readme_52c8cceceba5.gif\" alt=\"sahi-yolox\">\u003C\u002Fa>\n\n### Framework Agnostic Sliced\u002FStandard Prediction\n\n\u003Cimg width=\"700\" alt=\"sahi-predict\" src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fobss_sahi_readme_00c75787f7e1.gif\">\n\nFind detailed info on using `sahi predict` command in the\n[CLI documentation](docs\u002Fcli.md#predict-command-usage) and explore the\n[prediction API](docs\u002Fpredict.md) for advanced usage.\n\nFind detailed info on video inference at\n[video inference tutorial](https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fdiscussions\u002F626).\n\n### Error Analysis Plots & Evaluation\n\n\u003Cimg width=\"700\" alt=\"sahi-analyse\" src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fobss_sahi_readme_254d79ad5201.gif\">\n\nFind detailed info at\n[Error Analysis Plots & Evaluation](https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fdiscussions\u002F622).\n\n### Interactive Visualization & Inspection\n\n\u003Cimg width=\"700\" alt=\"sahi-fiftyone\" src=\"https:\u002F\u002Fuser-images.githubusercontent.com\u002F34196005\u002F149321540-e6dd5f3-36dc-4267-8574-a985dd0c6578.gif\">\n\nExplore [FiftyOne integration](docs\u002Ffiftyone.md) for interactive visualization\nand inspection.\n\n### Other Utilities\n\nCheck the [comprehensive COCO utilities guide](docs\u002Fcoco.md) for YOLO\nconversion, dataset slicing, subsampling, filtering, merging, and splitting\noperations. Learn more about the [slicing utilities](docs\u002Fslicing.md) for\ndetailed control over image and dataset slicing parameters.\n\n## \u003Cdiv align=\"center\">Citation\u003C\u002Fdiv>\n\nIf you use this package in your work, please cite as:\n\n```bibtex\n@article{akyon2022sahi,\n title={Slicing Aided Hyper Inference and Fine-tuning for Small Object Detection},\n author={Akyon, Fatih Cagatay and Altinuc, Sinan Onur and Temizel, Alptekin},\n journal={2022 IEEE International Conference on Image Processing (ICIP)},\n doi={10.1109\u002FICIP46576.2022.9897990},\n pages={966-970},\n year={2022}\n}\n```\n\n```bibtex\n@software{obss2021sahi,\n author = {Akyon, Fatih Cagatay and Cengiz, Cemil and Altinuc, Sinan Onur and Cavusoglu, Devrim and Sahin, Kadir and Eryuksel, Ogulcan},\n title = {{SAHI: A lightweight vision library for performing large scale object detection and instance segmentation}},\n month = nov,\n year = 2021,\n publisher = {Zenodo},\n doi = {10.5281\u002Fzenodo.5718950},\n url = {https:\u002F\u002Fdoi.org\u002F10.5281\u002Fzenodo.5718950}\n}\n```\n\n## \u003Cdiv align=\"center\">Contributing\u003C\u002Fdiv>\n\nWe welcome contributions! Please see our [Contributing Guide](CONTRIBUTING.md)\nto get started. Thank you 🙏 to all our contributors!\n\n\u003Cp align=\"center\">\n \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fgraphs\u002Fcontributors\">\n \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fobss_sahi_readme_c63ad92911bf.png\" \u002F>\n \u003C\u002Fa>\n\u003C\u002Fp>\n","\u003Cdiv align=\"center\">\n\u003Ch1>\n SAHI：切片辅助超推理\n\u003C\u002Fh1>\n\n\u003Ch4>\n 一个用于大规模目标检测与实例分割的轻量级视觉库\n\u003C\u002Fh4>\n\n\u003Ch4>\n \u003Cimg width=\"700\" alt=\"teaser\" src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fobss_sahi_readme_97f7e16a88ae.png\">\n\u003C\u002Fh4>\n\n\u003C!-- 下载与版本 -->\n\u003Cdiv>\n \u003Ca href=\"https:\u002F\u002Fpepy.tech\u002Fproject\u002Fsahi\">\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fobss_sahi_readme_fae08394318e.png\" alt=\"总下载量\">\u003C\u002Fa>\n \u003Ca href=\"https:\u002F\u002Fpepy.tech\u002Fproject\u002Fsahi\">\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fobss_sahi_readme_fae08394318e.png\u002Fmonth\" alt=\"月度下载量\">\u003C\u002Fa>\n \u003Ca href=\"https:\u002F\u002Fbadge.fury.io\u002Fpy\u002Fsahi\">\u003Cimg src=\"https:\u002F\u002Fbadge.fury.io\u002Fpy\u002Fsahi.svg\" alt=\"PyPI版本\">\u003C\u002Fa>\n \u003Ca href=\"https:\u002F\u002Fanaconda.org\u002Fconda-forge\u002Fsahi\">\u003Cimg src=\"https:\u002F\u002Fanaconda.org\u002Fconda-forge\u002Fsahi\u002Fbadges\u002Fversion.svg\" alt=\"Conda版本\">\u003C\u002Fa>\n \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fblob\u002Fmain\u002FLICENSE.md\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fpypi\u002Fl\u002Fsahi\" alt=\"许可证\">\u003C\u002Fa>\n\u003C\u002Fdiv>\n\n\u003C!-- CI与质量 -->\n\u003Cdiv>\n \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Factions\u002Fworkflows\u002Fci.yml\">\u003Cimg src=\"https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Factions\u002Fworkflows\u002Fci.yml\u002Fbadge.svg\" alt=\"CI\">\u003C\u002Fa>\n \u003Ca href=\"https:\u002F\u002Fsecurity.snyk.io\u002Fpackage\u002Fpip\u002Fsahi\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FSnyk_security-monitored-8A2BE2\" alt=\"已知漏洞\">\u003C\u002Fa>\n \u003Ca href=\"https:\u002F\u002Fwww.codefactor.io\u002Frepository\u002Fgithub\u002Fonuralpszr\u002Fsahi\">\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fobss_sahi_readme_9ee0cb95ac54.png\" alt=\"CodeFactor\">\u003C\u002Fa>\n \u003Ca href=\"https:\u002F\u002Fieeexplore.ieee.org\u002Fdocument\u002F9897990\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FDOI-10.1109%2FICIP46576.2022.9897990-orange.svg\" alt=\"DOI\">\u003C\u002Fa>\n\u003C\u002Fdiv>\n\n\u003C!-- AI与文档 -->\n\u003Cdiv>\n \u003Ca href=\"https:\u002F\u002Fcontext7.com\u002Fobss\u002Fsahi\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FContext7%20MCP-Indexed-blue\" alt=\"Context7 MCP\">\u003C\u002Fa>\n \u003Ca href=\"https:\u002F\u002Fcontext7.com\u002Fobss\u002Fsahi\u002Fllms.txt\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fllms.txt-✓-brightgreen\" alt=\"llms.txt\">\u003C\u002Fa>\n \u003Ca href=\"https:\u002F\u002Fdeepwiki.com\u002Fobss\u002Fsahi\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FDeepWiki-obss%2Fsahi-blue.svg?logo=data:image\u002Fpng;base64,iVBORw0KGgoAAAANSUhEUgAAACwAAAAyCAYAAAAnWDnqAAAAAXNSR0IArs4c6QAAA05JREFUaEPtmUtyEzEQhtWTQyQLHNak2AB7ZnyXZMEjXMGeK\u002FAIi+QuHrMnbChYY7MIh8g01fJoopFbEqqcbWTp06\u002Fuv1saEDv4O3n3dV60RfP947Mm9\u002FSQc0ICFQgzfc4CYZoTPAswgSJCCUJUnAAoRHOAUOcATwbmVLWdGoH\u002F\u002FPB8mnKqScAhsD0kYP3j\u002FYt5LPQe2KvcXmGvRHcDnpxfL2zOYJ1mFwrryWTz0advv1Ut4CJgf5uhDuDj5eUcAUoahrdY\u002F56ebRWeraTjMt\u002F00Sh3UDtjgHtQNHwcRGOC98BJEAEymycmYcWwOprTgcB6VZ5JK5TAJ+fXGLBm3FDAmn6oPPjR4rKCAoJCal2eAiQp2x0vxTPB3ALO2CRkwmDy5WohzBDwSEFKRwPbknEggCPB\u002FimwrycgxX2NzoMCHhPkDwqYMr9tRcP5qNrMZHkVnOjRMWwLCcr8ohBVb1OMjxLwGCvjTikrsBOiA6fNyCrm8V1rP93iVPpwaE+gO0SsWmPiXB+jikdf6SizrT5qKasx5j8ABbHpFTx+vFXp9EnYQmLx02h1QTTrl6eDqxLnGjporxl3NL3agEvXdT0WmEost648sQOYAeJS9Q7bfUVoMGnjo4AZdUMQku50McDcMWcBPvr0SzbTAFDfvJqwLzgxwATnCgnp4wDl6Aa+Ax283gghmj+vj7feE2KBBRMW3FzOpLOADl0Isb5587h\u002FU4gGvkt5v60Z1VLG8BhYjbzRwyQZemwAd6cCR5\u002FXFWLYZRIMpX39AR0tjaGGiGzLVyhse5C9RKC6ai42ppWPKiBagOvaYk8lO7DajerabOZP46Lby5wKjw1HCRx7p9sVMOWGzb\u002FvA1hwiWc6jm3MvQDTogQkiqIhJV0nBQBTU+3okKCFDy9WwferkHjtxib7t3xIUQtHxnIwtx4mpg26\u002FHfwVNVDb4oI9RHmx5WGelRVlrtiw43zboCLaxv46AZeB3IlTkwouebTr1y2NjSpHz68WNFjHvupy3q8TFn3Hos2IAk4Ju5dCo8B3wP7VPr\u002FFGaKiG+T+v+TQqIrOqMTL1VdWV1DdmcbO8KXBz6esmYWYKPwDL5b5FA1a0hwapHiom0r\u002FcKaoqr+27\u002FXcrS5UwSMbQAAAABJRU5ErkJggg==\" alt=\"DeepWiki\">\u003C\u002Fa>\n \u003Ca href=\"https:\u002F\u002Fhuggingface.co\u002Fspaces\u002Ffcakyon\u002Fsahi-yolox\">\u003Cimg src=\"https:\u002F\u002Fraw.githubusercontent.com\u002Fobss\u002Fsahi\u002Fmain\u002Fresources\u002Fhf_spaces_badge.svg\" alt=\"HuggingFace Spaces\">\u003C\u002Fa>\n\u003C\u002Fdiv>\n\n\u003C\u002Fdiv>\n\n## \u003Cdiv align=\"center\">概述\u003C\u002Fdiv>\n\nSAHI 通过支持在大图像中检测小物体的 **切片推理**，帮助开发者克服目标检测中的现实挑战。它支持多种流行的检测模型，并提供易于使用的 API。\n\n\u003Cdiv align=\"center\">\n\n🌐 [英文](README.md) | 🇨🇳 [简体中文](docs\u002Fzh\u002FREADME.md)\n\n\u003C\u002Fdiv>\n\n| 命令 | 描述 |\n| ----------------------------------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |\n| [predict](https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fblob\u002Fmain\u002Fdocs\u002Fcli.md#predict-command-usage) | 使用任何 [ultralytics](https:\u002F\u002Fgithub.com\u002Fultralytics\u002Fultralytics) \u002F [mmdet](https:\u002F\u002Fgithub.com\u002Fopen-mmlab\u002Fmmdetection) \u002F [huggingface](https:\u002F\u002Fhuggingface.co\u002Fmodels?pipeline_tag=object-detection&sort=downloads) \u002F [torchvision](https:\u002F\u002Fpytorch.org\u002Fvision\u002Fstable\u002Fmodels.html#object-detection) 模型执行切片或标准视频\u002F图像预测 — 请参阅 [CLI 指南](docs\u002Fcli.md#predict-command-usage) |\n| [predict-fiftyone](https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fblob\u002Fmain\u002Fdocs\u002Fcli.md#predict-fiftyone-command-usage) | 使用任何受支持的模型执行切片或标准预测，并在 [fiftyone 应用程序](https:\u002F\u002Fgithub.com\u002Fvoxel51\u002Ffiftyone) 中探索结果 — [了解更多](docs\u002Ffiftyone.md) |\n| [coco slice](https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fblob\u002Fmain\u002Fdocs\u002Fcli.md#coco-slice-command-usage) | 自动切分 COCO 注释和图像文件 — 请参阅 [切片工具](docs\u002Fslicing.md) |\n| [coco fiftyone](https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fblob\u002Fmain\u002Fdocs\u002Fcli.md#coco-fiftyone-command-usage) | 使用 [fiftyone UI](https:\u002F\u002Fgithub.com\u002Fvoxel51\u002Ffiftyone) 按误检数量排序，探索您的 COCO 数据集上的多个预测结果 |\n| [coco evaluate](https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fblob\u002Fmain\u002Fdocs\u002Fcli.md#coco-evaluate-command-usage) | 对给定的预测和真实标签评估按类别划分的 COCO AP 和 AR — 请查看 [COCO 工具](docs\u002Fcoco.md) |\n| [coco analyse](https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fblob\u002Fmain\u002Fdocs\u002Fcli.md#coco-analyse-command-usage) | 计算并导出多种错误分析图表 — 请参阅[完整指南](docs\u002FREADME.md) |\n| [coco yolo](https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fblob\u002Fmain\u002Fdocs\u002Fcli.md#coco-yolo-command-usage) | 自动将任何 COCO 数据集转换为 [ultralytics](https:\u002F\u002Fgithub.com\u002Fultralytics\u002Fultralytics) 格式 |\n\n### 社区认可\n\n[📜 引用 SAHI 的出版物列表（目前超过 600 篇）](https:\u002F\u002Fscholar.google.com\u002Fscholar?hl=en&as_sdt=2005&sciodt=0,5&cites=14065474760484865747&scipsc=&q=&scisbd=1)\n\n[🏆 使用过 SAHI 的竞赛获奖者列表](https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fdiscussions\u002F688)\n\n### AI 工具认可\n\nSAHI 的文档已被\n[indexed in Context7 MCP](https:\u002F\u002Fcontext7.com\u002Fobss\u002Fsahi)，为 AI 编码助手提供最新、特定于版本的代码示例和 API 参考。我们还提供一个遵循新兴 AI 可读文档标准的 [llms.txt](https:\u002F\u002Fcontext7.com\u002Fobss\u002Fsahi\u002Fllms.txt) 文件。要将 SAHI 文档集成到您的 AI 开发工作流中，请查看\n[Context7 MCP 安装指南](https:\u002F\u002Fgithub.com\u002Fupstash\u002Fcontext7#%EF%B8%8F-installation)。\n\n## \u003Cdiv align=\"center\">安装\u003C\u002Fdiv>\n\n### 基本安装\n\n```bash\npip install sahi\n```\n\n\u003Cdetails closed>\n\u003Csummary>\n\u003Cbig>\u003Cb>详细安装（点击展开）\u003C\u002Fb>\u003C\u002Fbig>\n\u003C\u002Fsummary>\n\n- 安装您所需的 PyTorch 和 torchvision 版本：\n\n```console\npip install torch==2.7.0 torchvision==0.22.0 --index-url https:\u002F\u002Fdownload.pytorch.org\u002Fwhl\u002Fcu126\n```\n\n（mmdet 支持需要 PyTorch 2.1.2）：\n\n```console\npip install torch==2.1.2 torchvision==0.16.2 --index-url https:\u002F\u002Fdownload.pytorch.org\u002Fwhl\u002Fcu121\n```\n\n- 安装您所需的检测框架（ultralytics）：\n\n```console\npip install ultralytics>=8.3.161\n```\n\n- 安装您所需的检测框架（huggingface）：\n\n```console\npip install transformers>=4.49.0 timm\n```\n\n- 安装您所需的检测框架（yolov5）：\n\n```console\npip install yolov5==7.0.14 sahi==0.11.21\n```\n\n- 安装您所需的检测框架（mmdet）：\n\n```console\npip install mim\nmim install mmdet==3.3.0\n```\n\n- 安装您所需的检测框架（roboflow）：\n\n```console\npip install inference>=0.51.5 rfdetr>=1.6.2\n```\n\n\u003C\u002Fdetails>\n\n## \u003Cdiv align=\"center\">快速入门\u003C\u002Fdiv>\n\n### 学习资源\n\n| 资源 | 类型 |\n| --------------------------------------------------------------------------------------------------------------------------------------------------- | ---------- |\n| [SAHI 简介](https:\u002F\u002Fmedium.com\u002Fcodable\u002Fsahi-a-vision-library-for-performing-sliced-inference-on-large-images-small-objects-c8b086af3b80) | 博文 |\n| [2025 年视频教程](https:\u002F\u002Fwww.youtube.com\u002Fwatch?v=ILqMBah5ZvI) ⭐ | 视频 |\n| [官方论文](https:\u002F\u002Fieeexplore.ieee.org\u002Fdocument\u002F9897990)（ICIP 2022 口头报告） | 论文 |\n| [预训练权重及 ICIP 2022 论文相关文件](https:\u002F\u002Fgithub.com\u002Ffcakyon\u002Fsmall-object-detection-benchmark) | 基准测试 |\n| [使用 FiftyOne 可视化和评估 SAHI 预测结果](https:\u002F\u002Fvoxel51.com\u002Fblog\u002Fhow-to-detect-small-objects\u002F) | 博文 |\n| [探索 SAHI – learnopencv.com](https:\u002F\u002Flearnopencv.com\u002Fslicing-aided-hyper-inference\u002F) | 文章 |\n| [Encord 解释的切片辅助超推理](https:\u002F\u002Fencord.com\u002Fblog\u002Fslicing-aided-hyper-inference-explained\u002F) | 文章 |\n| [视频教程：SAHI 用于小目标检测](https:\u002F\u002Fwww.youtube.com\u002Fwatch?v=UuOJKxn-M8&t=270s) | 视频 |\n| [卫星图像中的目标检测](https:\u002F\u002Fblog.ml6.eu\u002Fhow-to-detect-small-objects-in-very-large-images-70234bab0f98) | 博文 |\n| [COCO 数据集转换](https:\u002F\u002Fmedium.com\u002Fcodable\u002Fconvert-any-dataset-to-coco-object-detection-format-with-sahi-95349e1fe2b7) | 博文 |\n| [Kaggle 笔记本](https:\u002F\u002Fwww.kaggle.com\u002Fremekkinas\u002Fsahi-slicing-aided-hyper-inference-yv5-and-yx) | 笔记本 |\n| [错误分析图表与评估](https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fdiscussions\u002F622) ⭐ | 讨论 |\n| [交互式结果可视化与检查](https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fdiscussions\u002F624) ⭐ | 讨论 |\n| [视频推理支持](https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fdiscussions\u002F626) | 讨论 |\n| [切片操作笔记本](demo\u002Fslicing.ipynb) | 笔记本 |\n| [完整文档](docs\u002FREADME.md) | 文档 |\n\n### 笔记本与演示\n\n| 框架 | 笔记本 | 演示 |\n| ------------------ | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------- |\n| YOLO12 | [![在 Colab 中打开](https:\u002F\u002Fcolab.research.google.com\u002Fassets\u002Fcolab-badge.svg)](https:\u002F\u002Fcolab.research.google.com\u002Fgithub\u002Fobss\u002Fsahi\u002Fblob\u002Fmain\u002Fdemo\u002Finference_for_ultralytics.ipynb) | — |\n| YOLO11 | [![在 Colab 中打开](https:\u002F\u002Fcolab.research.google.com\u002Fassets\u002Fcolab-badge.svg)](https:\u002F\u002Fcolab.research.google.com\u002Fgithub\u002Fobss\u002Fsahi\u002Fblob\u002Fmain\u002Fdemo\u002Finference_for_ultralytics.ipynb) | — |\n| YOLO11-OBB | [![在 Colab 中打开](https:\u002F\u002Fcolab.research.google.com\u002Fassets\u002Fcolab-badge.svg)](https:\u002F\u002Fcolab.research.google.com\u002Fgithub\u002Fobss\u002Fsahi\u002Fblob\u002Fmain\u002Fdemo\u002Finference_for_ultralytics.ipynb) | — |\n| Roboflow \u002F RF-DETR | [![在 Colab 中打开](https:\u002F\u002Fcolab.research.google.com\u002Fassets\u002Fcolab-badge.svg)](https:\u002F\u002Fcolab.research.google.com\u002Fgithub\u002Fobss\u002Fsahi\u002Fblob\u002Fmain\u002Fdemo\u002Finference_for_roboflow.ipynb) | — |\n| RT-DETR v2 | [![在 Colab 中打开](https:\u002F\u002Fcolab.research.google.com\u002Fassets\u002Fcolab-badge.svg)](https:\u002F\u002Fcolab.research.google.com\u002Fgithub\u002Fobss\u002Fsahi\u002Fblob\u002Fmain\u002Fdemo\u002Finference_for_huggingface.ipynb) | — |\n| RT-DETR | [![在 Colab 中打开](https:\u002F\u002Fcolab.research.google.com\u002Fassets\u002Fcolab-badge.svg)](https:\u002F\u002Fcolab.research.google.com\u002Fgithub\u002Fobss\u002Fsahi\u002Fblob\u002Fmain\u002Fdemo\u002Finference_for_rtdetr.ipynb) | — |\n| HuggingFace | [![在 Colab 中打开](https:\u002F\u002Fcolab.research.google.com\u002Fassets\u002Fcolab-badge.svg)](https:\u002F\u002Fcolab.research.google.com\u002Fgithub\u002Fobss\u002Fsahi\u002Fblob\u002Fmain\u002Fdemo\u002Finference_for_huggingface.ipynb) | — |\n| YOLOv5 | [![在 Colab 中打开](https:\u002F\u002Fcolab.research.google.com\u002Fassets\u002Fcolab-badge.svg)](https:\u002F\u002Fcolab.research.google.com\u002Fgithub\u002Fobss\u002Fsahi\u002Fblob\u002Fmain\u002Fdemo\u002Finference_for_yolov5.ipynb) | — |\n| MMDetection | [![在 Colab 中打开](https:\u002F\u002Fcolab.research.google.com\u002Fassets\u002Fcolab-badge.svg)](https:\u002F\u002Fcolab.research.google.com\u002Fgithub\u002Fobss\u002Fsahi\u002Fblob\u002Fmain\u002Fdemo\u002Finference_for_mmdetection.ipynb) | — |\n| TorchVision | [![在 Colab 中打开](https:\u002F\u002Fcolab.research.google.com\u002Fassets\u002Fcolab-badge.svg)](https:\u002F\u002Fcolab.research.google.com\u002Fgithub\u002Fobss\u002Fsahi\u002Fblob\u002Fmain\u002Fdemo\u002Finference_for_torchvision.ipynb) | — |\n| YOLOX | — | [![HuggingFace Spaces](https:\u002F\u002Fraw.githubusercontent.com\u002Fobss\u002Fsahi\u002Fmain\u002Fresources\u002Fhf_spaces_badge.svg)](https:\u002F\u002Fhuggingface.co\u002Fspaces\u002Ffcakyon\u002Fsahi-yolox) |\n\n\u003Ca href=\"https:\u002F\u002Fhuggingface.co\u002Fspaces\u002Ffcakyon\u002Fsahi-yolox\">\u003Cimg width=\"600\" src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fobss_sahi_readme_52c8cceceba5.gif\" alt=\"sahi-yolox\">\u003C\u002Fa>\n\n### 框架无关的切片\u002F标准预测\n\n\u003Cimg width=\"700\" alt=\"sahi-predict\" src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fobss_sahi_readme_00c75787f7e1.gif\">\n\n有关使用 `sahi predict` 命令的详细信息，请参阅\n[CLI 文档](docs\u002Fcli.md#predict-command-usage)，并探索\n[预测 API](docs\u002Fpredict.md) 以获取高级用法。\n\n有关视频推理的详细信息，请参阅\n[视频推理教程](https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fdiscussions\u002F626)。\n\n### 错误分析图表与评估\n\n\u003Cimg width=\"700\" alt=\"sahi-analyse\" src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fobss_sahi_readme_254d79ad5201.gif\">\n\n详细信息请参阅\n[错误分析图表与评估](https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fdiscussions\u002F622)。\n\n### 交互式可视化与检查\n\n\u003Cimg width=\"700\" alt=\"sahi-fiftyone\" src=\"https:\u002F\u002Fuser-images.githubusercontent.com\u002F34196005\u002F149321540-e6dd5f3-36dc-4267-8574-a985dd0c6578.gif\">\n\n探索 [FiftyOne 集成](docs\u002Ffiftyone.md)，以实现交互式可视化和检查。\n\n### 其他工具\n\n请查阅[全面的 COCO 工具指南](docs\u002Fcoco.md)，了解 YOLO 转换、数据集切片、子采样、过滤、合并和拆分等操作。更多关于[切片工具](docs\u002Fslicing.md)的信息，可帮助您对图像和数据集的切片参数进行精细控制。\n\n## \u003Cdiv align=\"center\">引用\u003C\u002Fdiv>\n\n如果您在工作中使用本软件包，请按以下格式引用：\n\n```bibtex\n@article{akyon2022sahi,\n title={Slicing Aided Hyper Inference and Fine-tuning for Small Object Detection},\n author={Akyon, Fatih Cagatay and Altinuc, Sinan Onur and Temizel, Alptekin},\n journal={2022 IEEE International Conference on Image Processing (ICIP)},\n doi={10.1109\u002FICIP46576.2022.9897990},\n pages={966-970},\n year={2022}\n}\n```\n\n```bibtex\n@software{obss2021sahi,\n author = {Akyon, Fatih Cagatay and Cengiz, Cemil and Altinuc, Sinan Onur and Cavusoglu, Devrim and Sahin, Kadir and Eryuksel, Ogulcan},\n title = {{SAHI: 用于大规模目标检测和实例分割的轻量级视觉库}},\n month = nov,\n year = 2021,\n publisher = {Zenodo},\n doi = {10.5281\u002Fzenodo.5718950},\n url = {https:\u002F\u002Fdoi.org\u002F10.5281\u002Fzenodo.5718950}\n}\n```\n\n## \u003Cdiv align=\"center\">贡献\u003C\u002Fdiv>\n\n我们欢迎各类贡献！请参阅我们的[贡献指南](CONTRIBUTING.md)以开始参与。感谢所有贡献者 🙏！\n\n\u003Cp align=\"center\">\n \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fgraphs\u002Fcontributors\">\n \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fobss_sahi_readme_c63ad92911bf.png\" \u002F>\n \u003C\u002Fa>\n\u003C\u002Fp>","# SAHI 快速上手指南\n\nSAHI (Slicing Aided Hyper Inference) 是一个轻量级视觉库，专为解决大图像中小目标检测难题而设计。它通过**切片推理**技术，将大图切割成小块分别检测，再合并结果，显著提升小目标的检测精度。\n\n## 1. 环境准备\n\n* **操作系统**: Linux, macOS, Windows\n* **Python**: 3.8+\n* **核心依赖**: PyTorch, torchvision\n* **可选检测框架**: Ultralytics (YOLOv8\u002Fv10), MMDetection, Hugging Face Transformers, YOLOv5 等\n\n> **注意**：请确保已安装与你的 CUDA 版本匹配的 PyTorch。如果使用 `mmdet`，需特定版本的 PyTorch (如 2.1.2)。\n\n## 2. 安装步骤\n\n### 基础安装\n直接通过 pip 安装核心库：\n\n```bash\npip install sahi\n```\n\n### 完整环境配置（推荐）\n根据你使用的检测模型框架，选择以下一种方案进行安装：\n\n**方案 A：使用 Ultralytics (YOLOv8\u002Fv10 等)**\n```bash\n# 安装 PyTorch (示例为 CUDA 12.6，请根据实际情况调整)\npip install torch==2.7.0 torchvision==0.22.0 --index-url https:\u002F\u002Fdownload.pytorch.org\u002Fwhl\u002Fcu126\n\n# 安装 Ultralytics\npip install ultralytics>=8.3.161\n\n# 安装 SAHI\npip install sahi\n```\n\n**方案 B：使用 Hugging Face Models**\n```bash\npip install torch torchvision\npip install transformers>=4.49.0 timm\npip install sahi\n```\n\n**方案 C：使用 MMDetection**\n```bash\npip install torch==2.1.2 torchvision==0.16.2 --index-url https:\u002F\u002Fdownload.pytorch.org\u002Fwhl\u002Fcu121\npip install mim\nmim install mmdet==3.3.0\npip install sahi\n```\n\n> **国内加速建议**：如遇下载缓慢，可使用清华或阿里镜像源。\n> 例如：`pip install sahi -i https:\u002F\u002Fpypi.tuna.tsinghua.edu.cn\u002Fsimple`\n\n## 3. 基本使用\n\nSAHI 提供了简单的 Python API 和命令行工具。以下是两种最常用的方式。\n\n### 方式一：Python API 调用（最灵活）\n\n适用于在代码中集成切片推理功能。\n\n```python\nfrom sahi import AutoDetectionModel\nfrom sahi.predict import get_sliced_prediction\n\n# 1. 加载检测模型 (以 YOLOv8 为例)\ndetection_model = AutoDetectionModel.from_pretrained(\n model_path='yolov8n.pt',\n confidence_threshold=0.25,\n device=\"cuda\", # 或 \"cpu\"\n)\n\n# 2. 执行切片推理\n# source: 图片路径、视频路径或目录\n# slice_height\u002Fwidth: 切片尺寸\n# overlap_height_ratio\u002Foverlap_width_ratio: 重叠率\nresult = get_sliced_prediction(\n \"large_image.jpg\",\n detection_model,\n slice_height=512,\n slice_width=512,\n overlap_height_ratio=0.2,\n overlap_width_ratio=0.2\n)\n\n# 3. 查看结果或导出\nprint(f\"检测到 {len(result.object_prediction_list)} 个目标\")\nresult.export_visuals(export_dir=\".\u002Foutput\")\n```\n\n### 方式二：命令行工具 (CLI)\n\n适用于快速测试或对批量数据进行处理，无需编写代码。\n\n**对单张图片进行切片预测：**\n```bash\nsahi predict \\\n --source image.jpg \\\n --model_path yolov8n.pt \\\n --model_type yolov8 \\\n --slice_height 512 \\\n --slice_width 512 \\\n --overlap_height_ratio 0.2 \\\n --overlap_width_ratio 0.2 \\\n --conf_th 0.25 \\\n --device cuda\n```\n\n**常用参数说明：**\n* `--model_type`: 支持 `yolov8`, `mmdet`, `huggingface`, `torchvision` 等。\n* `--slice_height` \u002F `--slice_width`: 切片的大小，小目标较多时可适当调小。\n* `--overlap_*_ratio`: 切片间的重叠比例，防止目标被切断，通常设为 0.2。\n* `--novisual`: 添加此参数可跳过可视化图片的生成，仅保存 JSON 结果。","某安防团队正在处理城市级监控视频，需要从数亿张高分辨率（如 4K\u002F8K）的航拍或广角截图中自动识别远处的行人和车辆。\n\n### 没有 sahi 时\n- **小目标漏检严重**：直接将整张大图输入模型，远处微小的行人因像素占比过低，被模型完全忽略，检出率不足 40%。\n- **显存溢出崩溃**：试图通过强行缩放图片来适配模型输入，导致图像细节模糊；若保持原图尺寸，则直接触发 GPU 显存溢出（OOM），程序频繁崩溃。\n- **调试黑盒难优化**：面对大量漏检数据，缺乏可视化的误差分析工具，开发人员只能盲目调整阈值，无法定位是切片策略问题还是模型本身缺陷。\n- **推理速度不可控**：为了兼顾精度尝试手动编写切片逻辑，代码耦合度高且推理耗时极长，无法满足实时性要求。\n\n### 使用 sahi 后\n- **精准捕获微小目标**：利用 sahi 的切片辅助推理技术，将大图自动切割重叠处理，远处行人的检出率提升至 90% 以上，彻底解决小目标丢失问题。\n- **资源消耗平稳可控**：无需修改原有模型架构，sahi 在内存中高效管理切片队列，既避免了显存溢出，又保持了原生模型的推理流畅度。\n- **可视化误差分析**：直接调用内置的交互式 UI 和误差分析图表，直观看到哪些区域漏检，快速迭代优化切片参数，排查效率提升数倍。\n- **框架无关即插即用**：无论是 YOLO 系列还是 Mask R-CNN，只需几行代码即可接入 sahi，自动并行化处理大幅缩短了从开发到部署的周期。\n\nsahi 通过智能切片技术，让现有检测模型在不重新训练的情况下，具备了在大规模高分辨率图像中精准识别微小目标的“超能力”。","https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fobss_sahi_52c8ccec.gif","obss","Open Business Software Solutions","https:\u002F\u002Foss.gittoolsai.com\u002Favatars\u002Fobss_94a7e0a3.png","Open Source for Open Business",null,"rcm@obss.tech","obsstech","https:\u002F\u002Fobss.tech","https:\u002F\u002Fgithub.com\u002Fobss",[82,86],{"name":83,"color":84,"percentage":85},"Python","#3572A5",99.5,{"name":87,"color":88,"percentage":89},"Shell","#89e051",0.5,5243,743,"2026-04-18T06:03:57","MIT","未说明","非必需（支持 CPU），若使用 GPU 需根据 PyTorch 版本匹配 CUDA（示例中提及 cu126, cu121）",{"notes":97,"python":94,"dependencies":98},"该库为轻量级视觉库，核心功能是切片推理。安装时需注意：不同检测框架（如 mmdet, yolov5）对 PyTorch 版本有特定要求（例如 mmdet 需要 torch==2.1.2）。支持多种后端包括 Ultralytics (YOLO), MMDetection, HuggingFace, TorchVision, Roboflow 等。可通过 pip 或 conda 安装。",[99,100,101,102,103,104,105,106,107,108],"torch","torchvision","ultralytics>=8.3.161","transformers>=4.49.0","timm","yolov5==7.0.14","mmdet==3.3.0","inference>=0.51.5","rfdetr>=1.6.2","fiftyone",[15,14],[111,112,113,114,115,116,117,118,119,120,121,122,123,108,124,125,126,127,128,129],"object-detection","instance-segmentation","computer-vision","small-object-detection","large-image","mmdetection","pytorch","python","coco","deep-learning","machine-learning","remote-sensing","huggingface","satellite","tiling","explainable-ai","oriented-object-detection","yolo11","hacktoberfest","2026-03-27T02:49:30.150509","2026-04-19T06:02:49.598564",[],[134,139,144,149,154,159,164,169,174,179,184,189,194,199,204,209,214,219,224,229],{"id":135,"version":136,"summary_zh":137,"released_at":138},333763,"0.11.36","## 变更内容\n\n本次发布修复了 CLI 中的 `torch` 导入错误。\n\n* 发布：✨ 在 `pyproject.toml` 中将 SAHI 版本升级至 0.11.35 (#1252)，由 @onuralpszr 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1252 中完成。\n\n\n**完整变更日志**：https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fcompare\u002F0.11.35...0.11.36","2025-09-28T22:20:30",{"id":140,"version":141,"summary_zh":142,"released_at":143},333764,"0.11.35","## 变更内容\n* 修复当前失败的 CI 流水线：通过添加 hatch 构建目标，由 @gboeer 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1245 中完成。\n* 重构：♻️ 提升代码库的可读性和类型提示，由 @onuralpszr 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1249 中完成。\n* 将 actions\u002Fsetup-python 从版本 5 升级到版本 6，由 @dependabot[bot] 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1244 中完成。\n* 重构：♻️ 更新 CI 工作流以支持多操作系统，并优化作业命名，由 @onuralpszr 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1250 中完成。\n* 修复因 shapely_annotation.multipolygon 为空而导致的空边界框问题，由 @vinnik-dmitry07 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1140 中完成。\n* 通过使用 shapely 中的 STRtree，显著提升后处理速度（NMS、NMM、GREEDYNMM），由 @nikvo1 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1248 中完成。\n* 修复：🐞 添加可选依赖项并改进 models 模块中的包管理，同时对 torch 进行延迟导入以修复 CI，由 @onuralpszr 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1251 中完成。\n\n## 新贡献者\n* @vinnik-dmitry07 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1140 中完成了首次贡献。\n* @nikvo1 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1248 中完成了首次贡献。\n\n**完整变更日志**：https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fcompare\u002F0.11.34...0.11.35","2025-09-26T19:43:31",{"id":145,"version":146,"summary_zh":147,"released_at":148},333765,"0.11.34","## 变更内容\n* 文档：添加 MkDocs Material 徽章，以表示支持，由 @onuralpszr 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1236 中完成。\n* 修复：在多 GPU 环境下，子进程中始终将模型加载到 “cuda:0” 上，由 @malopez00 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1230 中完成。\n* 重构：更新类别重新映射逻辑，使其适用于不可变类别，由 @gboeer 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1234 中完成。\n* 功能：添加对 torchvision Mask R-CNN 实例分割模型的支持，由 @curtiskennedy 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1238 中完成。\n* 重构：默认启用图像读取时的 EXIF 旋转，并使用 PIL 内置的 ImageOps 模块，由 @fcakyon 和 @onuralpszr 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1239 中完成。\n* 构建：版本号从 0.11.33 升级至 0.11.34，由 @fcakyon 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1241 中完成。\n\n## 新贡献者\n* @malopez00 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1230 中完成了首次贡献。\n* @curtiskennedy 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1238 中完成了首次贡献。\n\n**完整变更日志**：https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fcompare\u002F0.11.33...0.11.34","2025-08-31T11:02:58",{"id":150,"version":151,"summary_zh":152,"released_at":153},333766,"0.11.33","## 变更内容\n* 文档：简化并优化 context7 配置，由 @fcakyon 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1218 中完成\n* 构建：消除不必要的 CI 工作流触发，由 @fcakyon 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1219 中完成\n* 修复：修正错误的 context7 配置结构，由 @fcakyon 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1220 中完成\n* 重构：🧹 改进测试和工具代码的组织结构，由 @onuralpszr 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1223 中完成\n* 👷 将 actions\u002Fcheckout 从 4 升级到 5，由 @dependabot[bot] 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1229 中完成\n* 修复：🐛 更新版本至 0.11.33，以修复开发环境中的 sahi 安装问题，由 @onuralpszr 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1233 中完成\n\n\n**完整变更日志**：https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fcompare\u002F0.11.32...0.11.33","2025-08-22T11:34:14",{"id":155,"version":156,"summary_zh":157,"released_at":158},333767,"0.11.32","## 变更内容\n* 文档：📝 由 @onuralpszr 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1210 中引入 mkdocs-material，为 sahi 项目添加文档页面\n* 开发：由 @dependabot[bot] 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1211 中将 astral-sh\u002Fsetup-uv 从 5 升级到 6\n* 修复：🐞 由 @onuralpszr 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1212 中将 publish_docs 工作流的部署分支从 'develop' 更新为 'main'\n* 文档：📝 由 @onuralpszr 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1213 中再次引入 mkdocs-material，为 sahi 项目添加文档页面\n* 在 `README.md` 中添加 `DeepWiki` 文档徽章，由 @RizwanMunawar 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1215 中完成\n* 重构：🛠 由 @onuralpszr 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1214 中移除冗余的日志导入，并将日志配置集中化\n* 版本更新至 0.11.32，由 @fcakyon 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1216 中完成\n\n## 新贡献者\n* @onuralpszr 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1210 中完成了首次贡献\n* @dependabot[bot] 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1211 中完成了首次贡献\n* @RizwanMunawar 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1215 中完成了首次贡献\n\n**完整变更日志**：https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fcompare\u002F0.11.31...0.11.32","2025-08-02T11:33:22",{"id":160,"version":161,"summary_zh":162,"released_at":163},333768,"0.11.31","## 变更内容\n* 使 Category 不可变，并由 @gboeer 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1206 中添加测试\n* 由 @kikefdezl 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1205 中更新 greedy_nmm 的文档字符串\n* 由 @fcakyon 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1208 中更新版本号\n\n## 新贡献者\n* @kikefdezl 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1205 中完成了首次贡献\n\n**完整变更日志**: https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fcompare\u002F0.11.30...0.11.31","2025-07-15T08:41:48",{"id":165,"version":166,"summary_zh":167,"released_at":168},333769,"0.11.30"," # SAHI v0.11.30 发行说明\n\n 我们很高兴地宣布 SAHI v0.11.30 正式发布！本次更新带来了更完善的性能跟踪、增强的测试基础设施以及更好的开发者体验！\n\n ## 📈 里程碑\n - **引用 SAHI 的学术论文已达到 400 篇！** ([#1168](https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1168))\n\n ## 🚀 主要更新\n\n ### 性能与监控\n - 修复了 `get_sliced_prediction` 中的后处理时长跟踪问题——现在能够准确区分切片、预测和后处理的耗时，从而实现更精确的性能监控\n ([#1201](https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1201)) - 感谢 @Toprak2！\n\n ### 框架支持更新\n - 重构了 Ultralytics 支持，新增 ONNX 模型支持并提升了兼容性\n ([#1184](https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1184))\n - 将 TorchVision 支持更新至最新 API\n ([#1182](https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1182))\n - 改进了 Detectron2 支持，优化了配置处理以避免 KeyError 问题\n ([#1116](https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1116)) - 感谢 @Arnesh1411！\n - 新增 Roboflow 框架支持，用于处理来自 Roboflow Universe 的 RF-DETR 模型\n ([#1161](https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1161)) - 感谢 @nok！\n - 移除了 deepsparse 集成，因为该框架已不再维护\n ([#1164](https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1164))\n\n ### 测试基础设施\n - 将测试套件迁移到 pytest\n ([#1187](https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1187))\n - 测试运行速度更快，支持更好的并行执行\n - 扩展了 Python 版本覆盖范围（3.8、3.9、3.10、3.11、3.12）\n - 更新至较新的 PyTorch 版本，以提升兼容性测试效果\n - 改进了测试组织结构，便于维护\n - 重构了 MMDetection 测试，以提高可靠性\n ([#1185](https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1185))\n\n ### 开发者体验\n - 新增 Context7 MCP 集成，助力 AI 辅助开发\n ([#1198](https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1198))\n - SAHI 文档现已在 Context7 MCP 中索引\n - 为 AI 编码助手提供最新且针对特定版本的代码示例\n - 包含 [llms.txt](https:\u002F\u002Fcontext7.com\u002Fobss\u002Fsahi\u002Fllms.txt) 文件，方便 AI 读取文档内容\n - 请参阅 [Context7 MCP 安装指南](https:\u002F\u002Fgithub.com\u002Fupstash\u002Fcontext7#%EF%B8%8F-installation)，将 SAHI 文档集成到您的 AI 工作流中\n\n ## 🛠️ 其他改进\n\n ### 代码质量与安全性\n - 引入不可变边界框，确保线程安全操作\n ([#1194](https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1194)、[#1191](https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1191)) - 感谢 @gboeer！\n - 在整个代码库中增强了类型提示和文档字符串\n ([#1195](https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1195)) - 感谢 @gboeer！\n - 为预测分数重载运算符，支持直观的分数比较\n ([#1190](https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1190)) - 感谢 @gboeer！\n - PyTorch 现已成为可选依赖项","2025-07-08T16:33:26",{"id":170,"version":171,"summary_zh":172,"released_at":173},333770,"0.11.29","## 变更内容\n* 由 @gboeer 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1194 中将边界框设为不可变\n* 由 @gboeer 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1195 中改进类型提示和文档字符串\n* 由 @fcakyon 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1196 中更新版本号\n\n\n**完整变更日志**: https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fcompare\u002F0.11.28...0.11.29","2025-07-04T09:45:35",{"id":175,"version":176,"summary_zh":177,"released_at":178},333771,"0.11.28","## 变更内容\n* @gboeer 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1190 中为预测分数添加了重载运算符\n* @Arnesh1411 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1116 中改进了 Detectron2 的支持\n* @gboeer 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1191 中对边界框使用不可变参数\n* @fcakyon 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1192 中更新了版本号\n\n## 新贡献者\n* @Arnesh1411 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1116 中做出了首次贡献\n\n**完整变更日志**: https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fcompare\u002F0.11.27...0.11.28","2025-07-01T19:17:43",{"id":180,"version":181,"summary_zh":182,"released_at":183},333772,"0.11.27","## 变更内容\n* 修复：将推理方法更新为使用“threshold”而非“confidence”，由 @nok 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1180 中完成\n* 更新 README.md，由 @nok 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1179 中完成\n* 改进 pyproject.toml，由 @fcakyon 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1181 中完成\n* 重构依赖管理及部分文档，由 @fcakyon 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1183 中完成\n* 更新：重构 ultralytics 支持，由 @fcakyon 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1184 中完成\n* 重构 mmdet 测试，由 @fcakyon 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1185 中完成\n* 将 torchvision 支持更新至最新 API，由 @fcakyon 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1182 中完成\n* 优化 mmdet 工作流触发条件，由 @fcakyon 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1186 中完成\n* 将测试迁移至 pytest，由 @fcakyon 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1187 中完成\n* 更新版本号，由 @fcakyon 在 https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1188 中完成\n\n\n**完整变更日志**：https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fcompare\u002F0.11.26...0.11.27","2025-06-30T17:23:44",{"id":185,"version":186,"summary_zh":187,"released_at":188},333773,"0.11.26","## What's Changed\r\n* Bump opencv packages from `4.10.0.84` to `4.11.0.86` by @ducviet00-h2 in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1171\r\n* Add new framework Roboflow (RFDETR models) by @nok in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1161\r\n* add new contributors to readme by @fcakyon in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1175\r\n* add roboflow+sahi colab url to readme by @fcakyon in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1177\r\n* update version by @fcakyon in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1176\r\n\r\n## New Contributors\r\n* @ducviet00-h2 made their first contribution in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1171\r\n* @nok made their first contribution in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1161\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fcompare\u002F0.11.25...0.11.26","2025-06-26T13:28:49",{"id":190,"version":191,"summary_zh":192,"released_at":193},333774,"0.11.25","## What's Changed\r\n* update sahi citation in readme by @fcakyon in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1168\r\n* remove matplotlib-stubs as its not maintained by @fcakyon in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1169\r\n* Fix torch import errors by @ducviet00 in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1172\r\n* update version by @fcakyon in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1173\r\n\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fcompare\u002F0.11.24...0.11.25","2025-06-24T09:36:44",{"id":195,"version":196,"summary_zh":197,"released_at":198},333775,"0.11.24","## What's Changed\r\n* Fix typo and scripts URL by @gboeer in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1155\r\n* fix ci workflow bug by @Dronakurl in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1156\r\n* [DOC] Fix typos by @gboeer in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1157\r\n* Remove deepsparse integration by @fcakyon in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1164\r\n* Fix: Make pytorch is not a hard dependency by @ducviet00 in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1162\r\n* fix: specify a device other than cuda:0 by @0xf21 in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1158\r\n* fix: correct regex string formatting in select_device function by @fcakyon in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1165\r\n* add TensorrtExecutionProvider to yolov8onnx by @p-constant in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1091\r\n* update version by @fcakyon in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1166\r\n\r\n## New Contributors\r\n* @gboeer made their first contribution in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1155\r\n* @ducviet00 made their first contribution in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1162\r\n* @0xf21 made their first contribution in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1158\r\n* @p-constant made their first contribution in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1091\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fcompare\u002F0.11.23...0.11.24","2025-06-22T08:38:28",{"id":200,"version":201,"summary_zh":202,"released_at":203},333776,"0.11.23","## What's Changed\r\n* fix(CI): numpy dependency fixes #1119 by @Dronakurl in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1144\r\n* Fix: Predict cannot find TIF files in source directory by @dibunker in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1142\r\n* Fixed typos in demo Notebooks by @picjul in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1150\r\n* fix: Fix Polygon Repair and Empty Polygon Issues, see #1118 by @mario-dg in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1138\r\n* improve package ci logging by @fcakyon in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1151\r\n\r\n## New Contributors\r\n* @dibunker made their first contribution in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1142\r\n* @picjul made their first contribution in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1150\r\n* @mario-dg made their first contribution in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1138\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fcompare\u002F0.11.22...0.11.23","2025-05-05T12:17:08",{"id":205,"version":206,"summary_zh":207,"released_at":208},333777,"0.11.22","## What's Changed\r\n* Improve suppot for latest mmdet (v3.3.0) by @fcakyon in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1129\r\n* Improve support for latest yolov5-pip and ultralytics versions by @fcakyon in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1130\r\n* support latest huggingface\u002Ftransformers models by @fcakyon in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1131\r\n* refctor coco to yolo conversion, update docs by @fcakyon in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1132\r\n* bump version by @fcakyon in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1134\r\n\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fcompare\u002F0.11.21...0.11.22\r\n\r\n## Core Documentation Files\r\n\r\n### [Prediction Utilities](predict.md)\r\n- Detailed guide for performing object detection inference\r\n- Standard and sliced inference examples\r\n- Batch prediction usage\r\n- Class exclusion during inference\r\n- Visualization parameters and export formats\r\n- Interactive examples with various model integrations (YOLOv8, MMDetection, etc.)\r\n\r\n### [Slicing Utilities](slicing.md)\r\n- Guide for slicing large images and datasets\r\n- Image slicing examples\r\n- COCO dataset slicing examples\r\n- Interactive demo notebook reference\r\n\r\n### [COCO Utilities](coco.md)\r\n- Comprehensive guide for working with COCO format datasets\r\n- Dataset creation and manipulation\r\n- Slicing COCO datasets\r\n- Dataset splitting (train\u002Fval)\r\n- Category filtering and updates\r\n- Area-based filtering\r\n- Dataset merging\r\n- Format conversion (COCO ↔ YOLO)\r\n- Dataset sampling utilities\r\n- Statistics calculation\r\n- Result validation\r\n\r\n### [CLI Commands](cli.md)\r\n- Complete reference for SAHI command-line interface\r\n- Prediction commands\r\n- FiftyOne integration\r\n- COCO dataset operations\r\n- Environment information\r\n- Version checking\r\n- Custom script usage\r\n\r\n### [FiftyOne Integration](fiftyone.md)\r\n- Guide for visualizing and analyzing predictions with FiftyOne\r\n- Dataset visualization\r\n- Result exploration\r\n- Interactive analysis\r\n\r\n## Interactive Examples\r\n\r\nAll documentation files are complemented by interactive Jupyter notebooks in the [demo directory](..\u002Fdemo\u002F):\r\n- `slicing.ipynb` - Slicing operations demonstration\r\n- `inference_for_ultralytics.ipynb` - YOLOv8\u002FYOLO11\u002FYOLO12 integration\r\n- `inference_for_yolov5.ipynb` - YOLOv5 integration\r\n- `inference_for_mmdetection.ipynb` - MMDetection integration\r\n- `inference_for_huggingface.ipynb` - HuggingFace models integration\r\n- `inference_for_torchvision.ipynb` - TorchVision models integration\r\n- `inference_for_rtdetr.ipynb` - RT-DETR integration\r\n- `inference_for_sparse_yolov5.ipynb` - DeepSparse optimized inference\r\n\r\n## Getting Started\r\n\r\nIf you're new to SAHI:\r\n\r\n1. Start with the [prediction utilities](predict.md) to understand basic inference\r\n2. Explore the [slicing utilities](slicing.md) to learn about processing large images\r\n3. Check out the [CLI commands](cli.md) for command-line usage\r\n4. Dive into [COCO utilities](coco.md) for dataset operations\r\n5. Try the interactive notebooks in the [demo directory](..\u002Fdemo\u002F) for hands-on experience\r\n","2025-03-09T00:08:38",{"id":210,"version":211,"summary_zh":212,"released_at":213},333778,"0.11.21","## What's Changed\r\n* Exclude classes from inference using pretrained or custom models by @gguzzy in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1104\r\n* pyproject.toml, pre-commit, ruff, uv and typing issues, fixes #1119 by @Dronakurl in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1120\r\n* add class exclusion example into predict docs by @gguzzy in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1125\r\n* Add OBB demo by @fcakyon in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1126\r\n* fix a type hint typo in predict func by @fcakyon in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1111\r\n* Remove numpy\u003C2 upper pin by @weiji14 in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1112\r\n* fix ci badge on readme by @fcakyon in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1124\r\n* fix version in pyproject.toml by @fcakyon in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1127\r\n\r\n## New Contributors\r\n* @Dronakurl made their first contribution in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1120\r\n* @gguzzy made their first contribution in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1104\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fcompare\u002F0.11.20...0.11.21","2025-03-06T05:17:26",{"id":215,"version":216,"summary_zh":217,"released_at":218},333779,"0.11.20","## What's Changed\r\n* add yolo11 and ultralytics obb task support by @fcakyon in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1109\r\n* support latest opencv version by @fcakyon in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1106\r\n* simplify yolo detection model code by @fcakyon in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1107\r\n* Pin shapely>2.0.0 by @weiji14 in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1101\r\n\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fcompare\u002F0.11.19...0.11.20","2024-12-16T17:07:26",{"id":220,"version":221,"summary_zh":222,"released_at":223},333780,"0.11.19","## What's Changed\r\n* fix ci actions by @fcakyon in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1073\r\n* Update has_mask method for mmdet models (handle an edge case) by @ccomkhj in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1066\r\n* Another self-intersection corner case handling by @sergiev in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F982\r\n* Update README.md by @fcakyon in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1077\r\n* drop non-working yolonas support by @fcakyon in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1097\r\n* drop yolonas support part2 by @fcakyon in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1098\r\n* Update has_mask method for mmdet models (handle ConcatDataset) by @ccomkhj in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1092\r\n\r\n## New Contributors\r\n* @ccomkhj made their first contribution in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1066\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fcompare\u002F0.11.18...0.11.19","2024-11-22T07:20:04",{"id":225,"version":226,"summary_zh":227,"released_at":228},333781,"0.11.18","## What's Changed\r\n* add yolov8 mask support, improve mask processing speed by 4-5x by @mayrajeo in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1039\r\n* fix has_mask method for mmdet models by @Alias-z in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1054\r\n* Fix `TypeError: 'GeometryCollection' object is not subscriptable` when slicing COCO by @Alias-z in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1047\r\n* support opencv-python version 4.9 by @iokarkan in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1041\r\n* add upperlimit to numpy dep by @fcakyon in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1057\r\n* add more unit tests by @MMerling in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1048\r\n* upgrade ci actions by @fcakyon in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1049\r\n\r\n## New Contributors\r\n* @iokarkan made their first contribution in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1041\r\n* @MMerling made their first contribution in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1048\r\n* @Alias-z made their first contribution in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1047\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fcompare\u002F0.11.16...0.11.18","2024-07-10T10:19:32",{"id":230,"version":231,"summary_zh":232,"released_at":233},333782,"0.11.16","## What's Changed\r\n* Update README.md by @fcakyon in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F966\r\n* Updated the constant variables for Yolov8 by @AmoghDhaliwal in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F917\r\n* Slicing: add unique slice index number to output file name by @jokober in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F943\r\n* update version by @fcakyon in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F967\r\n* Correcting type hints for `get_slice_bboxes()`. by @S-aiueo32 in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F930\r\n* Correcting `slice_image()` by @S-aiueo32 in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F931\r\n* Customize YOLOv8 image_size & device + Allow Saving Slices by @lakshaymehra in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F929\r\n* fix package testing workflow paths by @fcakyon in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F968\r\n* remove detectron2 from package tests by @fcakyon in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F979\r\n* ONNX runtime support by @karl-joan in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F922\r\n* Fix RLE when segmentation is None by @bobyard-com in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F996\r\n* update functions docstrings and type hinting by @fcakyon in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1016\r\n* fix postprocess type options description by @williamlung in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1013\r\n* revert back package version by @fcakyon in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1017\r\n* Improve printout readability by @jacobmarks in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1009\r\n* improve readme by @fcakyon in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1018\r\n* remove an unused list in postprocess by @developer0hye in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1002\r\n* add more contributors to readme by @fcakyon in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1019\r\n* add more contributors to readme by @fcakyon in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1020\r\n* Adds link to new FiftyOne tutorial by @jacobmarks in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1023\r\n* RTDETR implementation by @edugzlez in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F940\r\n* Improve yolov8 config by @GuillaumeBruand in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F988\r\n* update readme by @fcakyon in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1034\r\n* relax opencv version by @fcakyon in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1035\r\n* disable slice export by default by @fcakyon in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1036\r\n* remove quality param in slice export due to errors by @fcakyon in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1037\r\n* fix for using bgr image in inference instead of rgb by @bilkosem in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1022\r\n\r\n## New Contributors\r\n* @AmoghDhaliwal made their first contribution in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F917\r\n* @S-aiueo32 made their first contribution in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F930\r\n* @lakshaymehra made their first contribution in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F929\r\n* @karl-joan made their first contribution in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F922\r\n* @bobyard-com made their first contribution in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F996\r\n* @williamlung made their first contribution in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1013\r\n* @jacobmarks made their first contribution in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1009\r\n* @developer0hye made their first contribution in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1002\r\n* @edugzlez made their first contribution in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F940\r\n* @GuillaumeBruand made their first contribution in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F988\r\n* @bilkosem made their first contribution in https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fpull\u002F1022\r\n\r\n**Full Changelog**: https:\u002F\u002Fgithub.com\u002Fobss\u002Fsahi\u002Fcompare\u002F0.11.15...0.11.16","2024-05-20T08:40:09"]