[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"similar-am15h--tflite_flutter_plugin":3,"tool-am15h--tflite_flutter_plugin":62},[4,18,26,36,46,54],{"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":42,"last_commit_at":43,"category_tags":44,"status":17},8272,"opencode","anomalyco\u002Fopencode","OpenCode 是一款开源的 AI 编程助手（Coding Agent），旨在像一位智能搭档一样融入您的开发流程。它不仅仅是一个代码补全插件，而是一个能够理解项目上下文、自主规划任务并执行复杂编码操作的智能体。无论是生成全新功能、重构现有代码，还是排查难以定位的 Bug，OpenCode 都能通过自然语言交互高效完成，显著减少开发者在重复性劳动和上下文切换上的时间消耗。\n\n这款工具专为软件开发者、工程师及技术研究人员设计，特别适合希望利用大模型能力来提升编码效率、加速原型开发或处理遗留代码维护的专业人群。其核心亮点在于完全开源的架构，这意味着用户可以审查代码逻辑、自定义行为策略，甚至私有化部署以保障数据安全，彻底打破了传统闭源 AI 助手的“黑盒”限制。\n\n在技术体验上，OpenCode 提供了灵活的终端界面（Terminal UI）和正在测试中的桌面应用程序，支持 macOS、Windows 及 Linux 全平台。它兼容多种包管理工具，安装便捷，并能无缝集成到现有的开发环境中。无论您是追求极致控制权的资深极客，还是渴望提升产出的独立开发者，OpenCode 都提供了一个透明、可信",144296,1,"2026-04-16T14:50:03",[13,45],"插件",{"id":47,"name":48,"github_repo":49,"description_zh":50,"stars":51,"difficulty_score":32,"last_commit_at":52,"category_tags":53,"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":55,"name":56,"github_repo":57,"description_zh":58,"stars":59,"difficulty_score":32,"last_commit_at":60,"category_tags":61,"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",[45,13,15,14],{"id":63,"github_repo":64,"name":65,"description_en":66,"description_zh":67,"ai_summary_zh":68,"readme_en":69,"readme_zh":70,"quickstart_zh":71,"use_case_zh":72,"hero_image_url":73,"owner_login":74,"owner_name":75,"owner_avatar_url":76,"owner_bio":77,"owner_company":78,"owner_location":79,"owner_email":78,"owner_twitter":78,"owner_website":78,"owner_url":80,"languages":81,"stars":120,"forks":121,"last_commit_at":122,"license":123,"difficulty_score":124,"env_os":125,"env_gpu":126,"env_ram":127,"env_deps":128,"category_tags":135,"github_topics":78,"view_count":32,"oss_zip_url":78,"oss_zip_packed_at":78,"status":17,"created_at":136,"updated_at":137,"faqs":138,"releases":167},9201,"am15h\u002Ftflite_flutter_plugin","tflite_flutter_plugin","TensorFlow Lite Flutter Plugin","tflite_flutter_plugin 是一款专为 Flutter 开发者打造的 TensorFlow Lite 插件，旨在让移动端和桌面端应用轻松集成机器学习能力。它通过直接绑定高效的 TFLite C API，帮助开发者在 Android、iOS、Windows、Mac 及 Linux 等多平台上快速运行机器学习模型推理，解决了跨平台应用中模型部署复杂、性能难以兼顾的痛点。\n\n这款工具特别适合需要为 App 添加图像识别、自然语言处理等智能功能的移动应用开发者。其核心优势在于不仅保持了与原生 Java 或 Swift API 相似的开发体验，降低了学习门槛，还提供了卓越的性能表现。它支持利用多线程以及各类硬件加速代理（如 Android 的 GPU 和 NNAPI，iOS 的 Metal 和 CoreML，以及桌面的 XNNPack），确保推理速度接近原生应用水平。此外，tflite_flutter_plugin 允许开发者在不同的隔离线程（Isolates）中运行推理任务，有效避免阻塞主线程导致的界面卡顿，从而保障流畅的用户体验。无论是希望快速验证原型的工程师，还是追求极致性","tflite_flutter_plugin 是一款专为 Flutter 开发者打造的 TensorFlow Lite 插件，旨在让移动端和桌面端应用轻松集成机器学习能力。它通过直接绑定高效的 TFLite C API，帮助开发者在 Android、iOS、Windows、Mac 及 Linux 等多平台上快速运行机器学习模型推理，解决了跨平台应用中模型部署复杂、性能难以兼顾的痛点。\n\n这款工具特别适合需要为 App 添加图像识别、自然语言处理等智能功能的移动应用开发者。其核心优势在于不仅保持了与原生 Java 或 Swift API 相似的开发体验，降低了学习门槛，还提供了卓越的性能表现。它支持利用多线程以及各类硬件加速代理（如 Android 的 GPU 和 NNAPI，iOS 的 Metal 和 CoreML，以及桌面的 XNNPack），确保推理速度接近原生应用水平。此外，tflite_flutter_plugin 允许开发者在不同的隔离线程（Isolates）中运行推理任务，有效避免阻塞主线程导致的界面卡顿，从而保障流畅的用户体验。无论是希望快速验证原型的工程师，还是追求极致性能的资深开发者，都能借助它灵活地调用任意 TensorFlow Lite 模型，构建 robust 的智能应用。"," \u003Cp align=\"center\">\n    \u003Cbr>\n    \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fam15h_tflite_flutter_plugin_readme_eed77bccb58a.png\"\u002F>\n    \u003C\u002Fbr>\n\u003C\u002Fp>\n\u003Cp align=\"center\">\n \n   \u003Ca href=\"https:\u002F\u002Fflutter.dev\">\n     \u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FPlatform-Flutter-02569B?logo=flutter\"\n       alt=\"Platform\" \u002F>\n   \u003C\u002Fa>\n   \u003Ca href=\"https:\u002F\u002Fpub.dartlang.org\u002Fpackages\u002Ftflite_flutter\">\n     \u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fpub\u002Fv\u002Ftflite_flutter.svg\"\n       alt=\"Pub Package\" \u002F>\n   \u003C\u002Fa>\n    \u003Ca href=\"https:\u002F\u002Fpub.dev\u002Fdocumentation\u002Ftflite_flutter\u002Flatest\u002Ftflite_flutter\u002Ftflite_flutter-library.html\">\n        \u003Cimg alt=\"Docs\" src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fam15h_tflite_flutter_plugin_readme_b1821ab427ad.png\">\n    \u003C\u002Fa>\n    \u003Ca href=\"https:\u002F\u002Fopensource.org\u002Flicenses\u002FApache-2.0\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FLicense-Apache%202.0-blue.svg\">\u003C\u002Fa>\n\n\n\u003C\u002Fa>\n\u003C\u002Fp>\n\n## Announcement\n\nUpdate: 26 April, 2023\n\nThe TensorFlow team has officially migrated this project to a [new repository](https:\u002F\u002Fgithub.com\u002Ftensorflow\u002Fflutter-tflite), deprecating this one. We will be focusing on getting the plugin to a stable and usable state to help our developers add robust machine learning features to their Flutter apps. PRs and contributions are more than welcome there, though please be mindful that this is a work in progress, so some things may be a bit broken for a bit :)\n\nWe do want to say a *huge* thank you to Amish for working on this initial plugin, and we're excited to keep it progressing.\n\nFeel free to reach out to me with questions until then.\n\nThanks!\n\n- ptruiz@google.com\n\n## Overview\n\nTensorFlow Lite Flutter plugin provides a flexible and fast solution for accessing TensorFlow Lite interpreter and performing inference. The API is similar to the TFLite Java and Swift APIs. It directly binds to TFLite C API making it efficient (low-latency). Offers acceleration support using NNAPI, GPU delegates on Android, Metal and CoreML delegates on iOS, and XNNPack delegate on Desktop platforms.\n\n\n## Key Features\n\n* Multi-platform Support for Android, iOS, Windows, Mac, Linux.\n* Flexibility to use any TFLite Model.\n* Acceleration using multi-threading and delegate support.\n* Similar structure as TensorFlow Lite Java API.\n* Inference speeds close to native Android Apps built using the Java API.\n* You can choose to use any TensorFlow version by building binaries locally.\n* Run inference in different isolates to prevent jank in UI thread.\n\n\n## (Important) Initial setup : Add dynamic libraries to your app\n\n### Android\n\n1. Place the script [install.sh](https:\u002F\u002Fgithub.com\u002Fam15h\u002Ftflite_flutter_plugin\u002Fblob\u002Fmaster\u002Finstall.sh) (Linux\u002FMac) or [install.bat](https:\u002F\u002Fgithub.com\u002Fam15h\u002Ftflite_flutter_plugin\u002Fblob\u002Fmaster\u002Finstall.bat) (Windows) at the root of your project.\n\n2. Execute `sh install.sh` (Linux) \u002F `install.bat` (Windows) at the root of your project to automatically download and place binaries at appropriate folders.\n\n   Note: *The binaries installed will **not** include support for `GpuDelegateV2` and `NnApiDelegate` however `InterpreterOptions().useNnApiForAndroid` can still be used.* \n\n3. Use **`sh install.sh -d`** (Linux) or **`install.bat -d`** (Windows) instead if you wish to use these `GpuDelegateV2` and `NnApiDelegate`.\n\nThese scripts install pre-built binaries based on latest stable tensorflow release. For info about using other tensorflow versions follow [instructions in wiki](https:\u002F\u002Fgithub.com\u002Fam15h\u002Ftflite_flutter_plugin\u002Fwiki\u002F). \n\n### iOS\n\n1. Download [`TensorFlowLiteC.framework`](https:\u002F\u002Fgithub.com\u002Fam15h\u002Ftflite_flutter_plugin\u002Freleases\u002Fdownload\u002Fv0.5.0\u002FTensorFlowLiteC.framework.zip). For building a custom version of tensorflow, follow [instructions in wiki](https:\u002F\u002Fgithub.com\u002Fam15h\u002Ftflite_flutter_plugin\u002Fwiki\u002F). \n2. Place the `TensorFlowLiteC.framework` in the pub-cache folder of this package.\n\n Pub-Cache folder location: [(ref)](https:\u002F\u002Fdart.dev\u002Ftools\u002Fpub\u002Fcmd\u002Fpub-get#the-system-package-cache)\n\n - `~\u002F.pub-cache\u002Fhosted\u002Fpub.dartlang.org\u002Ftflite_flutter-\u003Cplugin-version>\u002Fios\u002F` (Linux\u002F Mac) \n - `%LOCALAPPDATA%\\Pub\\Cache\\hosted\\pub.dartlang.org\\tflite_flutter-\u003Cplugin-version>\\ios\\` (Windows)\n\n### Desktop\n\nFollow instructions in [this guide](https:\u002F\u002Fgithub.com\u002Fam15h\u002Ftflite_flutter_plugin\u002Fwiki\u002FBuilding-Desktop-binaries-with-XNNPack-Delegate) to build and use desktop binaries.\n\n## TFLite Flutter Helper Library\n\nA dedicated library with simple architecture for processing and manipulating input and output of TFLite Models. API design and documentation is identical to the TensorFlow Lite Android Support Library. Strongly recommended to be used with `tflite_flutter_plugin`. [Learn more](https:\u002F\u002Fgithub.com\u002Fam15h\u002Ftflite_flutter_helper). \n\n## Examples\n\n|Title|Code|Demo|Blog|\n|-----|----|----|----|\n|Text Classification App| [Code](https:\u002F\u002Fgithub.com\u002Fam15h\u002Ftflite_flutter_plugin\u002Ftree\u002Fmaster\u002Fexample)|\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fam15h_tflite_flutter_plugin_readme_4ed967edc4a5.gif\" width=120\u002F> |[Blog\u002FTutorial](https:\u002F\u002Fmedium.com\u002F@am15hg\u002Ftext-classification-using-tensorflow-lite-plugin-for-flutter-3b92f6655982)| \n|Image Classification App| [Code](https:\u002F\u002Fgithub.com\u002Fam15h\u002Ftflite_flutter_helper\u002Ftree\u002Fmaster\u002Fexample\u002Fimage_classification)|\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fam15h_tflite_flutter_plugin_readme_eada5b9bd467.gif\" width=120\u002F> |-|\n|Object Detection App| [Code](https:\u002F\u002Fgithub.com\u002Fam15h\u002Fobject_detection_flutter)|\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fam15h_tflite_flutter_plugin_readme_07968d1db950.gif\" width=120\u002F> |[Blog\u002FTutorial](https:\u002F\u002Fmedium.com\u002F@am15hg\u002Freal-time-object-detection-using-new-tensorflow-lite-flutter-support-ea41263e801d)|\n|Reinforcement Learning App| [Code](https:\u002F\u002Fgithub.com\u002Fwindmaple\u002Fplanestrike-flutter)|\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fam15h_tflite_flutter_plugin_readme_5094d26fd9e4.gif\" width=120\u002F> |[Blog\u002FTutorial](https:\u002F\u002Fwindmaple.medium.com\u002Fplaying-a-board-game-on-device-using-tensorflow-lite-and-fluter-a7c865b9aefc)| \n\n## Import\n\n    import 'package:tflite_flutter\u002Ftflite_flutter.dart';\n\n## Usage instructions\n\n### Creating the Interpreter\n\n* **From asset**\n\n    Place `your_model.tflite` in `assets` directory. Make sure to include assets in `pubspec.yaml`.\n\n    ```dart\n    final interpreter = await tfl.Interpreter.fromAsset('your_model.tflite');\n    ```\n\nRefer to the documentation for info on creating interpreter from buffer or file.\n\n### Performing inference\n\nSee [TFLite Flutter Helper Library](https:\u002F\u002Fwww.github.com\u002Fam15h\u002Ftflite_flutter_helper) for easy processing of input and output.\n\n* **For single input and output**\n\n    Use `void run(Object input, Object output)`.\n    ```dart\n    \u002F\u002F For ex: if input tensor shape [1,5] and type is float32\n    var input = [[1.23, 6.54, 7.81. 3.21, 2.22]];\n\n    \u002F\u002F if output tensor shape [1,2] and type is float32\n    var output = List.filled(1*2, 0).reshape([1,2]);\n\n    \u002F\u002F inference\n    interpreter.run(input, output);\n\n    \u002F\u002F print the output\n    print(output);\n    ```\n  \n* **For multiple inputs and outputs**\n\n    Use `void runForMultipleInputs(List\u003CObject> inputs, Map\u003Cint, Object> outputs)`.\n\n    ```dart\n    var input0 = [1.23];  \n    var input1 = [2.43];  \n\n    \u002F\u002F input: List\u003CObject>\n    var inputs = [input0, input1, input0, input1];  \n\n    var output0 = List\u003Cdouble>.filled(1, 0);  \n    var output1 = List\u003Cdouble>.filled(1, 0);\n\n    \u002F\u002F output: Map\u003Cint, Object>\n    var outputs = {0: output0, 1: output1};\n\n    \u002F\u002F inference  \n    interpreter.runForMultipleInputs(inputs, outputs);\n\n    \u002F\u002F print outputs\n    print(outputs)\n    ```\n\n### Closing the interpreter\n\n```dart\ninterpreter.close();\n```\n\n### Improve performance using delegate support\n\n    Note: This feature is under testing and could be unstable with some builds and on some devices.\n\n* **NNAPI delegate for Android**\n\n    ```dart\n    var interpreterOptions = InterpreterOptions()..useNnApiForAndroid = true;\n    final interpreter = await Interpreter.fromAsset('your_model.tflite',\n        options: interpreterOptions);\n\n    ```\n\n    or\n\n    ```dart\n    var interpreterOptions = InterpreterOptions()..addDelegate(NnApiDelegate());\n    final interpreter = await Interpreter.fromAsset('your_model.tflite',\n        options: interpreterOptions);\n\n    ```\n\n* **GPU delegate for Android and iOS**\n\n  * **Android** GpuDelegateV2\n\n    ```dart\n    final gpuDelegateV2 = GpuDelegateV2(\n            options: GpuDelegateOptionsV2(\n            false,\n            TfLiteGpuInferenceUsage.fastSingleAnswer,\n            TfLiteGpuInferencePriority.minLatency,\n            TfLiteGpuInferencePriority.auto,\n            TfLiteGpuInferencePriority.auto,\n        ));\n\n    var interpreterOptions = InterpreterOptions()..addDelegate(gpuDelegateV2);\n    final interpreter = await Interpreter.fromAsset('your_model.tflite',\n        options: interpreterOptions);\n    ```\n\n  * **iOS** Metal Delegate (GpuDelegate)\n\n    ```dart\n    final gpuDelegate = GpuDelegate(\n          options: GpuDelegateOptions(true, TFLGpuDelegateWaitType.active),\n        );\n    var interpreterOptions = InterpreterOptions()..addDelegate(gpuDelegate);\n    final interpreter = await Interpreter.fromAsset('your_model.tflite',\n        options: interpreterOptions);\n    ```\n\nRefer [Tests](https:\u002F\u002Fgithub.com\u002Fam15h\u002Ftflite_flutter_plugin\u002Fblob\u002Fmaster\u002Fexample\u002Fintegration_test\u002Ftflite_flutter_test.dart) to see more example code for each method.\n\n## Credits\n\n* Tian LIN, Jared Duke, Andrew Selle, YoungSeok Yoon, Shuangfeng Li from the TensorFlow Lite Team for their invaluable guidance.\n* Authors of [dart-lang\u002Ftflite_native](https:\u002F\u002Fgithub.com\u002Fdart-lang\u002Ftflite_native).\n","\u003Cp align=\"center\">\n    \u003Cbr>\n    \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fam15h_tflite_flutter_plugin_readme_eed77bccb58a.png\"\u002F>\n    \u003C\u002Fbr>\n\u003C\u002Fp>\n\u003Cp align=\"center\">\n \n   \u003Ca href=\"https:\u002F\u002Fflutter.dev\">\n     \u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FPlatform-Flutter-02569B?logo=flutter\"\n       alt=\"Platform\" \u002F>\n   \u003C\u002Fa>\n   \u003Ca href=\"https:\u002F\u002Fpub.dartlang.org\u002Fpackages\u002Ftflite_flutter\">\n     \u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fpub\u002Fv\u002Ftflite_flutter.svg\"\n       alt=\"Pub Package\" \u002F>\n   \u003C\u002Fa>\n    \u003Ca href=\"https:\u002F\u002Fpub.dev\u002Fdocumentation\u002Ftflite_flutter\u002Flatest\u002Ftflite_flutter\u002Ftflite_flutter-library.html\">\n        \u003Cimg alt=\"Docs\" src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fam15h_tflite_flutter_plugin_readme_b1821ab427ad.png\">\n    \u003C\u002Fa>\n    \u003Ca href=\"https:\u002F\u002Fopensource.org\u002Flicenses\u002FApache-2.0\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FLicense-Apache%202.0-blue.svg\">\u003C\u002Fa>\n\n\n\u003C\u002Fa>\n\u003C\u002Fp>\n\n## 公告\n\n更新：2023年4月26日\n\nTensorFlow 团队已正式将该项目迁移到一个新的仓库 [tensorflow\u002Fflutter-tflite](https:\u002F\u002Fgithub.com\u002Ftensorflow\u002Fflutter-tflite)，并废弃了此仓库。我们将专注于使该插件达到稳定且可用的状态，以帮助开发者在 Flutter 应用中添加强大的机器学习功能。欢迎在此新仓库中提交 PR 和贡献，但请注意，目前仍处于开发阶段，因此某些功能可能会暂时存在问题 :)\n\n我们非常感谢 Amish 为这个初始插件所做的工作，并期待它能够继续发展。\n\n在此期间，如有任何问题，欢迎随时与我联系。\n\n谢谢！\n\n- ptruiz@google.com\n\n## 概述\n\nTensorFlow Lite Flutter 插件提供了一种灵活且快速的解决方案，用于访问 TensorFlow Lite 解释器并执行推理。其 API 与 TFLite 的 Java 和 Swift API 类似。它直接绑定到 TFLite C API，因此效率高（低延迟）。支持使用 NNAPI、Android 上的 GPU 委托、iOS 上的 Metal 和 CoreML 委托，以及桌面平台上的 XNNPack 委托来加速推理过程。\n\n\n## 主要特性\n\n* 支持多平台：Android、iOS、Windows、Mac、Linux。\n* 可灵活使用任何 TFLite 模型。\n* 通过多线程和委托支持实现加速。\n* 结构与 TensorFlow Lite Java API 相似。\n* 推理速度接近使用 Java API 构建的原生 Android 应用。\n* 您可以通过本地构建二进制文件来选择使用任何 TensorFlow 版本。\n* 在不同的 Isolate 中运行推理，以避免阻塞 UI 线程。\n\n\n## （重要）初始设置：将动态库添加到您的应用\n\n### Android\n\n1. 将脚本 [install.sh](https:\u002F\u002Fgithub.com\u002Fam15h\u002Ftflite_flutter_plugin\u002Fblob\u002Fmaster\u002Finstall.sh)（Linux\u002FMac）或 [install.bat](https:\u002F\u002Fgithub.com\u002Fam15h\u002Ftflite_flutter_plugin\u002Fblob\u002Fmaster\u002Finstall.bat)（Windows）放置在项目的根目录下。\n\n2. 在项目根目录下执行 `sh install.sh`（Linux）或 `install.bat`（Windows），以自动下载并将二进制文件放置到相应文件夹中。\n\n   注意：*安装的二进制文件**不**包含对 `GpuDelegateV2` 和 `NnApiDelegate` 的支持，但仍然可以使用 `InterpreterOptions().useNnApiForAndroid`。*\n\n3. 如果您希望使用这些 `GpuDelegateV2` 和 `NnApiDelegate`，请改用 **`sh install.sh -d`**（Linux）或 **`install.bat -d`**（Windows）。\n\n这些脚本会根据最新的稳定版 TensorFlow 安装预编译的二进制文件。如需使用其他版本的 TensorFlow，请参阅 [wiki 中的说明](https:\u002F\u002Fgithub.com\u002Fam15h\u002Ftflite_flutter_plugin\u002Fwiki\u002F)。\n\n### iOS\n\n1. 下载 [`TensorFlowLiteC.framework`](https:\u002F\u002Fgithub.com\u002Fam15h\u002Ftflite_flutter_plugin\u002Freleases\u002Fdownload\u002Fv0.5.0\u002FTensorFlowLiteC.framework.zip)。如需构建自定义版本的 TensorFlow，请参阅 [wiki 中的说明](https:\u002F\u002Fgithub.com\u002Fam15h\u002Ftflite_flutter_plugin\u002Fwiki\u002F)。\n2. 将 `TensorFlowLiteC.framework` 放入此包的 pub-cache 文件夹中。\n\nPub-Cache 文件夹位置：[(参考)](https:\u002F\u002Fdart.dev\u002Ftools\u002Fpub\u002Fcmd\u002Fpub-get#the-system-package-cache)\n\n - `~\u002F.pub-cache\u002Fhosted\u002Fpub.dartlang.org\u002Ftflite_flutter-\u003Cplugin-version>\u002Fios\u002F`（Linux\u002FMac）\n - `%LOCALAPPDATA%\\Pub\\Cache\\hosted\\pub.dartlang.org\\tflite_flutter-\u003Cplugin-version>\\ios\\`（Windows）\n\n### 桌面\n\n请按照 [此指南](https:\u002F\u002Fgithub.com\u002Fam15h\u002Ftflite_flutter_plugin\u002Fwiki\u002FBuilding-Desktop-binaries-with-XNNPack-Delegate) 中的说明构建并使用桌面端的二进制文件。\n\n## TFLite Flutter 辅助库\n\n一个具有简单架构的专用库，用于处理和操作 TFLite 模型的输入和输出。其 API 设计和文档与 TensorFlow Lite Android 支持库完全相同。强烈建议与 `tflite_flutter_plugin` 一起使用。[了解更多信息](https:\u002F\u002Fgithub.com\u002Fam15h\u002Ftflite_flutter_helper)。\n\n## 示例\n\n|标题|代码|演示|博客|\n|-----|----|----|----|\n|文本分类应用| [代码](https:\u002F\u002Fgithub.com\u002Fam15h\u002Ftflite_flutter_plugin\u002Ftree\u002Fmaster\u002Fexample)|\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fam15h_tflite_flutter_plugin_readme_4ed967edc4a5.gif\" width=120\u002F> |[博客\u002F教程](https:\u002F\u002Fmedium.com\u002F@am15hg\u002Ftext-classification-using-tensorflow-lite-plugin-for-flutter-3b92f6655982)| \n|图像分类应用| [代码](https:\u002F\u002Fgithub.com\u002Fam15h\u002Ftflite_flutter_helper\u002Ftree\u002Fmaster\u002Fexample\u002Fimage_classification)|\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fam15h_tflite_flutter_plugin_readme_eada5b9bd467.gif\" width=120\u002F> |-|\n|目标检测应用| [代码](https:\u002F\u002Fgithub.com\u002Fam15h\u002Fobject_detection_flutter)|\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fam15h_tflite_flutter_plugin_readme_07968d1db950.gif\" width=120\u002F> |[博客\u002F教程](https:\u002F\u002Fmedium.com\u002F@am15hg\u002Freal-time-object-detection-using-new-tensorflow-lite-flutter-support-ea41263e801d)|\n|强化学习应用| [代码](https:\u002F\u002Fgithub.com\u002Fwindmaple\u002Fplanestrike-flutter)|\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fam15h_tflite_flutter_plugin_readme_5094d26fd9e4.gif\" width=120\u002F> |[博客\u002F教程](https:\u002F\u002Fwindmaple.medium.com\u002Fplaying-a-board-game-on-device-using-tensorflow-lite-and-fluter-a7c865b9aefc)| \n\n## 导入\n\n    import 'package:tflite_flutter\u002Ftflite_flutter.dart';\n\n## 使用说明\n\n### 创建解释器\n\n* **从资源文件加载**\n\n    将 `your_model.tflite` 放入 `assets` 目录中。确保在 `pubspec.yaml` 中包含这些资源。\n\n    ```dart\n    final interpreter = await tfl.Interpreter.fromAsset('your_model.tflite');\n    ```\n\n有关如何从缓冲区或文件创建解释器的信息，请参阅文档。\n\n### 执行推理\n\n有关输入和输出的便捷处理，请参阅 [TFLite Flutter 助手库](https:\u002F\u002Fwww.github.com\u002Fam15h\u002Ftflite_flutter_helper)。\n\n* **对于单个输入和输出**\n\n    使用 `void run(Object input, Object output)`。\n    ```dart\n    \u002F\u002F 例如：如果输入张量形状为 [1,5]，类型为 float32\n    var input = [[1.23, 6.54, 7.81, 3.21, 2.22]];\n\n    \u002F\u002F 如果输出张量形状为 [1,2]，类型为 float32\n    var output = List.filled(1*2, 0).reshape([1,2]);\n\n    \u002F\u002F 推理\n    interpreter.run(input, output);\n\n    \u002F\u002F 打印输出\n    print(output);\n    ```\n  \n* **对于多个输入和输出**\n\n    使用 `void runForMultipleInputs(List\u003CObject> inputs, Map\u003Cint, Object> outputs)`。\n\n    ```dart\n    var input0 = [1.23];  \n    var input1 = [2.43];  \n\n    \u002F\u002F 输入：List\u003CObject>\n    var inputs = [input0, input1, input0, input1];  \n\n    var output0 = List\u003Cdouble>.filled(1, 0);  \n    var output1 = List\u003Cdouble>.filled(1, 0);\n\n    \u002F\u002F 输出：Map\u003Cint, Object>\n    var outputs = {0: output0, 1: output1};\n\n    \u002F\u002F 推理  \n    interpreter.runForMultipleInputs(inputs, outputs);\n\n    \u002F\u002F 打印输出\n    print(outputs)\n    ```\n\n### 关闭解释器\n\n```dart\ninterpreter.close();\n```\n\n### 使用委托支持提升性能\n\n    注意：此功能目前仍在测试中，在某些构建版本和设备上可能不稳定。\n\n* **适用于 Android 的 NNAPI 委托**\n\n    ```dart\n    var interpreterOptions = InterpreterOptions()..useNnApiForAndroid = true;\n    final interpreter = await Interpreter.fromAsset('your_model.tflite',\n        options: interpreterOptions);\n\n    ```\n\n    或者\n\n    ```dart\n    var interpreterOptions = InterpreterOptions()..addDelegate(NnApiDelegate());\n    final interpreter = await Interpreter.fromAsset('your_model.tflite',\n        options: interpreterOptions);\n\n    ```\n\n* **适用于 Android 和 iOS 的 GPU 委托**\n\n  * **Android** GpuDelegateV2\n\n    ```dart\n    final gpuDelegateV2 = GpuDelegateV2(\n            options: GpuDelegateOptionsV2(\n            false,\n            TfLiteGpuInferenceUsage.fastSingleAnswer,\n            TfLiteGpuInferencePriority.minLatency,\n            TfLiteGpuInferencePriority.auto,\n            TfLiteGpuInferencePriority.auto,\n        ));\n\n    var interpreterOptions = InterpreterOptions()..addDelegate(gpuDelegateV2);\n    final interpreter = await Interpreter.fromAsset('your_model.tflite',\n        options: interpreterOptions);\n    ```\n\n  * **iOS** Metal 委托（GpuDelegate）\n\n    ```dart\n    final gpuDelegate = GpuDelegate(\n          options: GpuDelegateOptions(true, TFLGpuDelegateWaitType.active),\n        );\n    var interpreterOptions = InterpreterOptions()..addDelegate(gpuDelegate);\n    final interpreter = await Interpreter.fromAsset('your_model.tflite',\n        options: interpreterOptions);\n    ```\n\n请参考 [测试用例](https:\u002F\u002Fgithub.com\u002Fam15h\u002Ftflite_flutter_plugin\u002Fblob\u002Fmaster\u002Fexample\u002Fintegration_test\u002Ftflite_flutter_test.dart)，以查看每种方法的更多示例代码。\n\n## 致谢\n\n* TensorFlow Lite 团队的 Tian LIN、Jared Duke、Andrew Selle、YoungSeok Yoon 和 Shuangfeng Li 提供了宝贵的指导。\n* [dart-lang\u002Ftflite_native](https:\u002F\u002Fgithub.com\u002Fdart-lang\u002Ftflite_native) 的作者们。","# tflite_flutter_plugin 快速上手指南\n\n## 环境准备\n\n*   **系统要求**：支持 Android、iOS、Windows、macOS 和 Linux。\n*   **前置依赖**：\n    *   已安装 Flutter SDK。\n    *   已配置对应平台的开发环境（如 Android Studio\u002FXcode）。\n*   **重要提示**：\n    > ⚠️ **官方公告**：TensorFlow 团队已将该项目的维护迁移至新仓库 [flutter-tflite](https:\u002F\u002Fgithub.com\u002Ftensorflow\u002Fflutter-tflite)。原仓库 (`am15h\u002Ftflite_flutter_plugin`) 已不再作为官方首选，但现有功能仍可使用。建议新项目优先考虑迁移后的官方版本，若需使用本指南内容，请知悉其处于维护过渡期。\n\n## 安装步骤\n\n### 1. 添加依赖\n在 `pubspec.yaml` 文件中添加插件：\n\n```yaml\ndependencies:\n  tflite_flutter: ^0.10.4 # 请使用最新可用版本\n```\n\n运行以下命令获取依赖：\n```bash\nflutter pub get\n```\n\n### 2. 配置动态库 (关键步骤)\n由于 TFLite 核心库需要原生二进制文件，必须执行以下脚本下载并放置库文件，否则应用无法运行。\n\n#### Android 配置\n在项目根目录下执行以下脚本（根据操作系统选择）：\n\n*   **Linux \u002F macOS**:\n    ```bash\n    sh install.sh\n    ```\n    *若需启用 GPU 加速 (GpuDelegateV2) 或 NNAPI，请使用:*\n    ```bash\n    sh install.sh -d\n    ```\n\n*   **Windows**:\n    ```bat\n    install.bat\n    ```\n    *若需启用 GPU 加速 (GpuDelegateV2) 或 NNAPI，请使用:*\n    ```bat\n    install.bat -d\n    ```\n\n> 注意：上述脚本会自动下载基于最新稳定版 TensorFlow 预编译的二进制文件。\n\n#### iOS 配置\n1.  下载 [`TensorFlowLiteC.framework`](https:\u002F\u002Fgithub.com\u002Fam15h\u002Ftflite_flutter_plugin\u002Freleases\u002Fdownload\u002Fv0.5.0\u002FTensorFlowLiteC.framework.zip)。\n2.  将解压后的 `TensorFlowLiteC.framework` 文件夹复制到插件的 iOS 目录中。\n    *   **Linux\u002FMac 路径**: `~\u002F.pub-cache\u002Fhosted\u002Fpub.dartlang.org\u002Ftflite_flutter-\u003C版本号>\u002Fios\u002F`\n    *   **Windows 路径**: `%LOCALAPPDATA%\\Pub\\Cache\\hosted\\pub.dartlang.org\\tflite_flutter-\u003C版本号>\\ios\\`\n\n#### Desktop (Windows\u002FMac\u002FLinux) 配置\n如需在桌面端使用并开启 XNNPack 加速，请参考官方 Wiki 构建对应的二进制文件。\n\n### 3. 推荐辅助库\n为了更方便地处理输入输出数据（如图片预处理、张量转换），强烈建议同时安装辅助库：\n\n```yaml\ndependencies:\n  tflite_flutter_helper: ^0.3.2 # 请检查最新版本\n```\n\n## 基本使用\n\n### 1. 导入包\n```dart\nimport 'package:tflite_flutter\u002Ftflite_flutter.dart';\n\u002F\u002F 推荐使用辅助库处理数据\nimport 'package:tflite_flutter_helper\u002Ftflite_flutter_helper.dart';\n```\n\n### 2. 加载模型\n将 `.tflite` 模型文件放入项目的 `assets` 目录，并在 `pubspec.yaml` 中注册：\n```yaml\nflutter:\n  assets:\n    - assets\u002Fyour_model.tflite\n```\n\n初始化解释器（Interpreter）：\n```dart\n\u002F\u002F 从资源文件加载\nfinal interpreter = await Interpreter.fromAsset('your_model.tflite');\n\n\u002F\u002F 可选：配置加速选项 (例如 Android NNAPI)\n\u002F\u002F var options = InterpreterOptions()..useNnApiForAndroid = true;\n\u002F\u002F final interpreter = await Interpreter.fromAsset('your_model.tflite', options: options);\n```\n\n### 3. 执行推理 (Inference)\n\n#### 场景 A：单输入单输出\n```dart\n\u002F\u002F 准备输入数据 (例如形状为 [1, 5] 的 float32 数据)\nvar input = [[1.23, 6.54, 7.81, 3.21, 2.22]];\n\n\u002F\u002F 准备输出缓冲区 (例如形状为 [1, 2])\nvar output = List.filled(1 * 2, 0.0).reshape([1, 2]);\n\n\u002F\u002F 运行推理\ninterpreter.run(input, output);\n\n\u002F\u002F 使用结果\nprint(output);\n```\n\n#### 场景 B：多输入多输出\n```dart\n\u002F\u002F 准备多个输入\nvar input0 = [1.23];\nvar input1 = [2.43];\nvar inputs = [input0, input1]; \n\n\u002F\u002F 准备多个输出容器\nvar output0 = List\u003Cdouble>.filled(1, 0);\nvar output1 = List\u003Cdouble>.filled(1, 0);\nvar outputs = {0: output0, 1: output1}; \u002F\u002F 映射索引到对象\n\n\u002F\u002F 运行推理\ninterpreter.runForMultipleInputs(inputs, outputs);\n\nprint(outputs);\n```\n\n### 4. 释放资源\n使用完毕后关闭解释器以释放内存：\n```dart\ninterpreter.close();\n```\n\n### 性能优化提示\n*   **隔离运行**：建议在独立的 isolate 中运行推理，避免阻塞 UI 线程导致卡顿。\n*   **硬件加速**：\n    *   **Android**: 支持 `NnApiDelegate` 和 `GpuDelegateV2`。\n    *   **iOS**: 支持 `Metal Delegate` (GpuDelegate) 和 `CoreML Delegate`。\n    *   **Desktop**: 支持 `XNNPack Delegate`。\n    *(具体配置代码请参考 Usage instructions 中的 Delegate 部分)*","一家初创团队正在开发一款跨平台健身指导 App，需要让用户通过手机摄像头实时识别深蹲动作是否标准。\n\n### 没有 tflite_flutter_plugin 时\n- 开发者必须分别为 Android 和 iOS 编写原生的 TensorFlow Lite 桥接代码，导致项目维护两套完全不同的底层逻辑，开发周期延长一倍。\n- 由于缺乏统一的隔离机制，复杂的模型推理直接阻塞了 UI 主线程，导致用户在扫码识别时界面频繁卡顿甚至无响应。\n- 无法便捷地调用 GPU 或 NNAPI 等硬件加速接口，模型在低端安卓机上推理延迟高达 500 毫秒，完全无法满足“实时”反馈的需求。\n- 每次更新 TensorFlow 版本都需要手动重新编译不同平台的动态库，极易因环境配置差异导致构建失败。\n\n### 使用 tflite_flutter_plugin 后\n- 团队仅需编写一套 Dart 代码即可同时部署到 Android、iOS 及桌面端，利用其类 Java API 结构快速集成姿态估计模型，上线时间缩短 60%。\n- 借助插件提供的 Isolate 支持，将耗时的推理任务移至后台线程运行，确保了即使在复杂计算下，App 界面依然保持 60fps 的流畅度。\n- 通过简单的配置即可启用 GPU Delegate 和 NNAPI，在普通手机上将单次推理耗时压缩至 30 毫秒以内，实现了真正的实时动作纠正。\n- 官方脚本自动管理各平台所需的二进制文件，开发者无需关心底层编译细节，轻松切换不同版本的 TensorFlow 内核。\n\ntflite_flutter_plugin 通过屏蔽底层异构差异并提供原生级性能，让 Flutter 开发者能像编写普通业务逻辑一样轻松落地高性能端侧 AI 应用。","https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fam15h_tflite_flutter_plugin_eed77bcc.png","am15h","Amish Garg","https:\u002F\u002Foss.gittoolsai.com\u002Favatars\u002Fam15h_8edb5d82.png","GSoC'20 & '21 @tensorflow | CS Undergrad, IIT Roorkee. ",null,"Roorkee","https:\u002F\u002Fgithub.com\u002Fam15h",[82,86,90,94,98,102,105,108,112,116],{"name":83,"color":84,"percentage":85},"Dart","#00B4AB",52,{"name":87,"color":88,"percentage":89},"C++","#f34b7d",26.8,{"name":91,"color":92,"percentage":93},"CMake","#DA3434",13.5,{"name":95,"color":96,"percentage":97},"Ruby","#701516",1.5,{"name":99,"color":100,"percentage":101},"Java","#b07219",1.4,{"name":103,"color":104,"percentage":101},"C","#555555",{"name":106,"color":107,"percentage":101},"Swift","#F05138",{"name":109,"color":110,"percentage":111},"Shell","#89e051",1.1,{"name":113,"color":114,"percentage":115},"Batchfile","#C1F12E",0.6,{"name":117,"color":118,"percentage":119},"Objective-C","#438eff",0.4,538,366,"2026-03-27T14:51:06","Apache-2.0",4,"Android, iOS, Windows, macOS, Linux","非必需。支持通过代理加速：Android (GPU Delegate V2, NNAPI), iOS (Metal\u002FCoreML), 桌面端 (XNNPack)。无需特定显卡型号或 CUDA 版本，依赖设备原生图形 API。","未说明",{"notes":129,"python":130,"dependencies":131},"1. 该项目已迁移至新仓库 (tensorflow\u002Fflutter-tflite)，当前仓库已弃用但仍可使用。\n2. Android 和 iOS 需手动运行脚本或下载 TensorFlowLiteC 二进制文件到指定目录。\n3. 若需使用 GPU 加速 (GpuDelegateV2) 或 NNAPI，Android 端需使用带 '-d' 参数的安装脚本。\n4. 桌面端 (Windows\u002FMac\u002FLinux) 需参考 Wiki 自行编译带有 XNNPack 代理的二进制文件。\n5. 建议在独立隔离区 (Isolate) 中运行推理以避免阻塞 UI 线程。","不适用 (基于 Flutter\u002FDart)",[132,133,134],"Flutter SDK","TensorFlowLiteC (二进制库)","tflite_flutter_helper (推荐)",[14,45],"2026-03-27T02:49:30.150509","2026-04-19T03:03:16.042658",[139,144,149,154,159,163],{"id":140,"question_zh":141,"answer_zh":142,"source_url":143},41317,"如何在 iOS 上解决 'Framework not found tflite_flutter' 或 Swift 库链接警告？","由于 `TensorFlowLiteC.framework` 二进制文件不再包含在仓库中，构建时可能会找不到框架。解决方案有两种：\n1. **本地副本法**：在项目中创建该包的本地副本，将预编译的二进制文件放入文件夹中，并在 `pubspec.yaml` 中引用本地包。\n2. **自动化脚本法（推荐用于 CI\u002FCD）**：在构建工具（如 Codemagic）的“预构建脚本”部分添加脚本，在构建前自动下载并解压框架到 `pub-cache`。示例脚本片段：\n```\nwget https:\u002F\u002Fgithub.com\u002Fam15h\u002Ftflite_flutter_plugin\u002Freleases\u002Fdownload\u002Fv0.5.0\u002FTensorFlowLiteC.framework.zip\nunzip TensorFlowLiteC.framework.zip\n# 将解压后的框架移动到正确位置\n```","https:\u002F\u002Fgithub.com\u002Fam15h\u002Ftflite_flutter_plugin\u002Fissues\u002F18",{"id":145,"question_zh":146,"answer_zh":147,"source_url":148},41318,"运行自定义 YOLO 模型时需要修改哪些文件？","除了修改 `Classifier` 文件中的 `MODEL_FILE_NAME`、`LABEL_FILE_NAME` 和 `INPUT_SIZE` 外，如果使用特定的辅助代码（如 TexMexMax 的示例），可能还需要修改 `tflite_flutter_helper` 包中的 `image_conversion.dart` 文件以适配模型输入格式。注意：修改辅助包可能导致版本冲突或构建错误，建议仔细核对版本兼容性（例如回退到 0.2.0 版本可能能运行但结果不准，需调整代码逻辑而非单纯降级）。","https:\u002F\u002Fgithub.com\u002Fam15h\u002Ftflite_flutter_plugin\u002Fissues\u002F158",{"id":150,"question_zh":151,"answer_zh":152,"source_url":153},41319,"遇到 'Bad state: failed precondition' 错误且涉及 Tensor 复制时如何解决？","该错误通常发生在 `tfLiteTensorCopyFromBuffer` 调用时，原因可能是输入张量的数据类型或形状与模型预期不匹配。请检查：\n1. 确保 `TensorImage` 的处理流程（如归一化 `NormalizeOp`）与模型训练时的预处理一致。\n2. 在新版本中如果找不到 `TensorImage`，可以使用 `package:image\u002Fimage.dart` 手动实现图像归一化到指定范围（例如 0-1 或 -1 到 1）。\n3. 验证输入缓冲区的字节长度是否与张量所需的字节数完全一致。","https:\u002F\u002Fgithub.com\u002Fam15h\u002Ftflite_flutter_plugin\u002Fissues\u002F167",{"id":155,"question_zh":156,"answer_zh":157,"source_url":158},41320,"项目维护状态如何？如果原维护者无法更新，社区该如何参与？","原仓库因维护者失联已趋于停滞。目前的建议是关注官方组织（tf org）下托管的新仓库，原仓库 README 已添加相关说明。社区成员若希望继续维护，不应仅仅创建克隆包导致生态分裂，而应尝试联系现有拥有写权限的成员合并 PR，或直接使用迁移后的官方新仓库以避免混淆。","https:\u002F\u002Fgithub.com\u002Fam15h\u002Ftflite_flutter_plugin\u002Fissues\u002F221",{"id":160,"question_zh":161,"answer_zh":162,"source_url":148},41321,"加载和关闭大型模型（如 24MB 的 YOLO 模型）时出现内存泄漏怎么办？","在频繁加载和关闭模型时确实可能观察到内存泄漏现象（尤其在 Xcode Instruments 中可见）。目前暂无官方修复的确切步骤，但建议开发者：\n1. 尽量复用 `Interpreter` 实例而不是反复创建和销毁。\n2. 监控应用内存使用情况，确认是否由模型本身引起还是其他资源未释放。\n3. 关注后续版本更新，社区正在讨论此问题。",{"id":164,"question_zh":165,"answer_zh":166,"source_url":153},41322,"如何在没有 `TensorImage` 类的新版本中执行图像归一化（NormalizeOp）？","在新版本中如果移除了 `TensorImage` 封装，可以直接使用 `package:image\u002Fimage.dart` 库手动处理图像。示例思路如下：\n```dart\nimport 'package:image\u002Fimage.dart' as imglib;\n\nimglib.Image normalize_image_into_range(imglib.Image image, num min, num max) {\n  \u002F\u002F 遍历像素并将值映射到 [min, max] 范围\n  \u002F\u002F 具体实现需根据模型要求的输入格式编写\n}\n```\n通过将图像像素值手动缩放到模型训练时使用的范围（如 0.0-1.0），可替代原有的 `NormalizeOp` 操作。",[168,173,177,182,187,191,196],{"id":169,"version":170,"summary_zh":171,"released_at":172},333255,"v0.9.0","- 支持 Windows、Mac 和 Linux 平台。 - 改进了 GPU 委托支持并修复了若干 bug。 - 支持 CoreML 和 XnnPack 委托。","2021-11-04T12:21:15",{"id":174,"version":175,"summary_zh":78,"released_at":176},333256,"tf_2.5","2021-06-20T21:20:10",{"id":178,"version":179,"summary_zh":180,"released_at":181},333257,"tf_2.4.1","\r\n","2021-04-06T18:41:46",{"id":183,"version":184,"summary_zh":185,"released_at":186},333258,"v0.5.0","适用于 tf.version=2.3.1 的 TensorFlow Lite 二进制文件\n\n","2020-07-17T10:42:55",{"id":188,"version":189,"summary_zh":78,"released_at":190},333259,"v0.4.0","2020-06-18T16:05:45",{"id":192,"version":193,"summary_zh":194,"released_at":195},333260,"v0.2.0","这些预构建的二进制文件基于 TensorFlow 2.2.0 稳定版。","2020-05-11T10:01:04",{"id":197,"version":198,"summary_zh":180,"released_at":199},333261,"v0.1.0","2020-04-09T19:31:17"]