Android 上的机器学习套件模型安装路径
使用集合让一切井井有条
根据您的偏好保存内容并对其进行分类。
默认情况下,所有 ML Kit 功能都使用 Google 训练的机器学习模型(称为“基础模型”)。本指南仅适用于基本模型。如需有关自定义模型的指南,请点击此处。
您可以通过以下三种方式之一安装 ML Kit API 中的模型:
- 未捆绑:模型通过 Google Play 服务下载和管理。
- 捆绑:模型在构建时会静态链接到您的应用。
- 动态下载:系统会按需下载模型。
每个 API 支持的安装路径
下表显示了每个 ML Kit 功能支持的模型安装路径:
API 专用指南会显示给定 API 可用的安装选项。
安装选项之间的主要区别
如何在捆绑和非捆绑之间进行选择
如果某个 API 同时支持捆绑和非捆绑安装选项,则:
如果您优先考虑以下事项,请使用捆绑选项:
- 在应用安装后立即完成功能
- 应用安装后在没有网络连接的情况下功能正常运行
如果您优先考虑以下事项,请使用未捆绑的选项:
- 缩减应用大小
- Google Play 服务自动更新模型
如何下载模型
使用未捆绑的模型选项时,您可以指定要将模型下载到设备的方式:
您可以通过向应用的 AndroidManifest.xml
文件添加声明来启用安装时模型下载。例如,以下代码段展示了如何配置应用,使之在从 Play 商店安装后自动下载条形码扫描模型:
<application ...>
...
<meta-data
android:name="com.google.mlkit.vision.DEPENDENCIES"
android:value="barcode" >
<!-- To use multiple models: android:value="barcode,model2,model3" -->
</application>
您可以通过 Google Play 服务 ModuleInstallClient API 请求明确下载。
如果您未启用在安装时下载模型的选项或未请求显式下载,系统将在您首次运行该功能时下载模型。在下载完成之前,推理请求将会失败。
如何更新模型
如需在使用捆绑的模型或动态下载的模型选项时更新模型,请执行以下操作:
更新应用的 Gradle 文件以使用最新的 ML Kit 功能客户端库。
dependencies {
implementation 'com.google.mlkit:barcode-scanning: 17.3.0' // The latest version number of the API
}
重新构建您的应用。
为何某些 API 提供动态下载的模型
某些 ML Kit API 包含太多的模型选项,无法打包。例如,数字手写识别支持 300 多种语言,但在安装过程中,并不一定需要将每种语言都添加到该功能中。为此,我们提供了第三种安装选项,即在安装后按需下载模型。目前,只有数字手写识别、翻译和实体提取支持此选项。
如未另行说明,那么本页面中的内容已根据知识共享署名 4.0 许可获得了许可,并且代码示例已根据 Apache 2.0 许可获得了许可。有关详情,请参阅 Google 开发者网站政策。Java 是 Oracle 和/或其关联公司的注册商标。
最后更新时间 (UTC):2025-08-29。
[null,null,["最后更新时间 (UTC):2025-08-29。"],[[["\u003cp\u003eML Kit features utilize Google-trained machine learning models, offering three installation options: unbundled, bundled, and dynamically downloaded.\u003c/p\u003e\n"],["\u003cp\u003eUnbundled models are downloaded via Google Play Services, while bundled models are included in the app, and dynamically downloaded models are fetched on demand.\u003c/p\u003e\n"],["\u003cp\u003eChoosing between bundled and unbundled models depends on priorities: bundled for immediate functionality, unbundled for smaller app size and automatic updates.\u003c/p\u003e\n"],["\u003cp\u003eModels can be downloaded at install time, explicitly requested, or downloaded on first use, and updates for bundled or dynamically downloaded models require app updates.\u003c/p\u003e\n"],["\u003cp\u003eDynamically downloaded models are offered for APIs with extensive model options, such as Digital Ink Recognition and Translation, to avoid bundling all possibilities.\u003c/p\u003e\n"]]],[],null,["All ML Kit features make use of Google-trained machine learning models (known as\nbase models) by default. This guide is only applicable to base models. [See here\nfor guidance on custom models.](/ml-kit/custom-models)\n\nModels in ML Kit APIs can be installed in one of three ways:\n\n1. **Unbundled:** Models are downloaded and managed via Google Play Services.\n2. **Bundled:** Models are statically linked to your app at build time.\n3. **Dynamically downloaded:** Models are downloaded on demand.\n\nSupported installation paths for each API\n\nThe table below shows which model installation paths are supported by each ML\nKit feature:\n\n|---------------------------------------------------------------------------|---------------|-------------|----------------------------|\n| | **Unbundled** | **Bundled** | **Dynamically downloaded** |\n| [Text recognition v2](/ml-kit/vision/text-recognition/v2/android) | ✅ | ✅ | |\n| [Face detection](/ml-kit/vision/face-detection/android) | ✅ | ✅ | |\n| [Face mesh detection](/ml-kit/vision/face-mesh-detection/android) | | ✅ | |\n| [Pose detection](/ml-kit/vision/pose-detection/android) | | ✅ | |\n| [Selfie segmentation](/ml-kit/vision/selfie-segmentation/android) | | ✅ | |\n| [Barcode scanning](/ml-kit/vision/barcode-scanning/android) | ✅ | ✅ | |\n| [Image labeling](/ml-kit/vision/image-labeling/android) | ✅ | ✅ | |\n| [Object detection and tracking](/ml-kit/vision/object-detection/android) | | ✅ | |\n| [Digital ink recognition](/ml-kit/vision/digital-ink-recognition/android) | | | ✅ |\n| [Document scanner](/ml-kit/vision/doc-scanner) | ✅ | | |\n| [Subject segmentation](/ml-kit/vision/subject-segmentation/android) | ✅ | | |\n| [Google code scanner](/ml-kit/code-scanner) | ✅ | | |\n| [Language identification](/ml-kit/language/identification/android) | ✅ | ✅ | |\n| [Translation](/ml-kit/language/translation/android) | | | ✅ |\n| [Smart Reply](/ml-kit/language/smart-reply/android) | ✅ | ✅ | |\n| [Entity Extraction](/ml-kit/language/entity-extraction/android) | | | ✅ |\n\nAPI-specific guides show which installation options are available for the given\nAPI.\n\nKey differences between installation options\n\n|----------------------------------------------|-------------------------------------------------------------------------------------------------------------------|----------------------------------------------------------------------------------------------------------------------------------------------|--------------------------------------------------------------------------------------------------------------------------------------------------------------------------|\n| | **Unbundled** | **Bundled** | **Dynamically downloaded** |\n| **Where are models stored?** | Resides in Google Play Services and is not counted towards the storage used by this app | Saved in [app-specific storage](https://developer.android.com/training/data-storage/app-specific) after installation | Saved in [app-specific storage](https://developer.android.com/training/data-storage/app-specific) after model download |\n| **How does model size affect app size?** | Does not contribute to app size | Directly contributes to app size | Does not contribute to app size but increase the [app-specific storage](https://developer.android.com/training/data-storage/app-specific) |\n| **When are models updated?** | Automatically updated when a newer version is released | [Need to update the app to update model](#update-model) | [Need to update the app to update model](#update-model) |\n| **When are models downloaded?** | [Models must be downloaded before use](#download-models) | All models and features are included when the app is installed, so are usable immediately | Model downloads, updates, and deletions must be managed manually using the [RemoteModelManager](/android/reference/com/google/mlkit/common/model/RemoteModelManager) API |\n| **When are models removed from the device?** | Google Play Services will only remove the model from storage if all apps that depend on the model are uninstalled | Models are removed from [app-specific storage](https://developer.android.com/training/data-storage/app-specific) when the app is uninstalled | Downloaded models are removed from [app-specific storage](https://developer.android.com/training/data-storage/app-specific) when the app is uninstalled |\n\nHow to choose between bundled and unbundled\n\nIf an API supports both the bundled and unbundled installation options:\n\n- Use the bundled option if you prioritize:\n\n - Complete feature functionality immediately after app installation\n - Feature functionality without network connection after app installation\n- Use the unbundled option if you prioritize:\n\n - Smaller app size\n - Automated model updates by Google Play Services\n\nHow to download models\n\nWhen using the unbundled model option, you can specify how you want models to be\ndownloaded to the device:\n\n- You can enable **install-time model downloads** by adding a declaration to\n your app's `AndroidManifest.xml` file. For example, the code snippet below\n shows how to configure your app to automatically download the Barcode\n Scanning model after your app is installed from the Play Store:\n\n \u003capplication ...\u003e\n ...\n \u003cmeta-data\n android:name=\"com.google.mlkit.vision.DEPENDENCIES\"\n android:value=\"barcode\" \u003e\n \u003c!-- To use multiple models: android:value=\"barcode,model2,model3\" --\u003e\n \u003c/application\u003e\n\n- You can **request explicit download** through Google Play Services\n [ModuleInstallClient API](/android/guides/module-install-apis).\n\n- If you don't enable install-time model downloads or request explicit\n download, the model will be downloaded the first time you run the feature.\n Until the download is completed, inference requests will fail.\n\nHow to update models\n\nTo update your models when using the bundled model or the dynamically downloaded\nmodels option:\n\n1. Update your app's gradle file to use the latest ML Kit feature client\n library.\n\n dependencies {\n implementation 'com.google.mlkit:barcode-scanning: 17.3.0' // The latest version number of the API\n }\n\n2. Rebuild your app.\n\nWhy some APIs offer dynamically downloaded models\n\nSome ML Kit API have too many model options to bundle. For example, [Digital ink\nrecognition](/ml-kit/vision/digital-ink-recognition) supports [300+\nlanguages](/ml-kit/vision/digital-ink-recognition/base-models), and it's not\nalways necessary to put every language inside the feature during installation.\nFor that purpose, we provide the third installation option, in which models are\ndownloaded on demand after installation. Currently, only [Digital ink\nrecognition](/ml-kit/vision/digital-ink-recognition),\n[Translation](/ml-kit/language/translation) and [Entity\nextraction](/ml-kit/language/entity-extraction) have this option."]]