google_mlkit_pose_detection 1.0.1

A Flutter plugin to use Google's ML Kit Pose Detection to detect the pose of a subject's body in real time from a continuous video or static image.

Google's ML Kit Pose Detection for Flutter

NOTE: This feature is still in Beta. Stay tuned for updates on Google's website.

A Flutter plugin to use Google's ML Kit Pose Detection to detect the pose of a subject's body in real time from a continuous video or static image.

PLEASE READ THIS before continuing or posting a new issue:

• Google's ML Kit was built only for mobile platforms: iOS and Android apps. Web or any other platform is not supported, you can request support for those platforms to Google in their repo.

• This plugin is not sponsored or maintained by Google. The authors are developers excited about Machine Learning that wanted to expose Google's native APIs to Flutter.

• Google's ML Kit APIs are only developed natively for iOS and Android. This plugin uses Flutter Platform Channels as explained here.

  Messages are passed between the client (the app/plugin) and host (platform) using platform channels as illustrated in this diagram:

  

  Messages and responses are passed asynchronously, to ensure the user interface remains responsive. To read more about platform channels go here.

  Because this plugin uses platform channels, no Machine Learning processing is done in Flutter/Dart, all the calls are passed to the native platform using MethodChannel in Android and FlutterMethodChannel in iOS, and executed using Google's native APIs. Think of this plugin as a bridge between your app and Google's native ML Kit APIs. This plugin only passes the call to the native API and the processing is done by Google's API. It is important that you understand this concept when it comes to debugging errors for your ML model and/or app.

• Since the plugin uses platform channels, you may encounter issues with the native API. Before submitting a new issue, identify the source of the issue. You can run both iOS and/or Android native example apps by Google and make sure that the issue is not reproducible with their native examples. If you can reproduce the issue in their apps then report the issue to Google. The authors do not have access to the source code of their native APIs, so you need to report the issue to them. If you find that their example apps are okay and still you have an issue using this plugin, then look at our closed and open issues. If you cannot find anything that can help you then report the issue and provide enough details. Be patient, someone from the community will eventually help you.

Requirements

iOS

• Minimum iOS Deployment Target: 15.0
• Xcode 15.3.0 or newer
• Swift 5.9 (Swift Package Manager)
• ML Kit does not support 32-bit architectures (i386 and armv7). ML Kit does support 64-bit architectures (x86_64 and arm64). Check this list to see if your device has the required device capabilities. More info here.

Swift Package Manager Setup

This plugin uses Swift Package Manager (SwiftPM) via d-date/google-mlkit-swiftpm (9.0.2), which also supports Apple Silicon iOS Simulators out of the box.

1. Linker Flags: In Xcode, go to Runner > Build Settings > Linking - General > Other Linker Flags (OTHER_LDFLAGS) and add:

   • -ObjC
   • -all_load

2. ML Kit Resource Bundles: SwiftPM cannot bundle model files inside .binaryTarget frameworks into the main app bundle automatically. Download MLKitPoseDetectionFastResources.bundle, MLKitPoseDetectionAccurateResources.bundle, MLKitPoseDetectionCommonResources.bundle, and MLKitXenoResources.bundle from the d-date/google-mlkit-swiftpm v9.0.2 release and add them to your Xcode Runner target (Build Phases > Copy Bundle Resources). See packages/example/ios/download_mlkit_bundles.sh for an example download script.

Note: While SwiftPM allows building and linking on Apple Silicon iOS Simulators, running Pose Detection inference at runtime requires a physical iOS device because Google's underlying MediaPipe graph requires device GPU/Metal support.

Android

• minSdkVersion: 21
• targetSdkVersion: 36
• compileSdkVersion: 36

⚠️ Production Disclaimer: Using this plugin in production is the responsibility of the developers consuming the plugin, not the authors. The authors provide this plugin as-is and are not responsible for any issues, failures, or compatibility problems that may arise from using this plugin in production environments.

Usage

Pose Detection

Create an instance of InputImage

Create an instance of InputImage as explained here.

final InputImage inputImage;

Create an instance of PoseDetector

final options = PoseDetectorOptions(
  mode: PoseDetectionMode.stream, // or PoseDetectionMode.single
  model: PoseDetectionModel.base, // or PoseDetectionModel.accurate
);
final poseDetector = PoseDetector(options: options);

Process image

final List<Pose> poses = await poseDetector.processImage(inputImage);

for (Pose pose in poses) {
  // to access all landmarks
  pose.landmarks.forEach((_, landmark) {
    final type = landmark.type;
    final x = landmark.x;
    final y = landmark.y;
    final z = landmark.z;
    final likelihood = landmark.likelihood;
  });

  // to access specific landmarks
  final landmark = pose.landmarks[PoseLandmarkType.nose];
}

Release resources with close()

poseDetector.close();

Example app

Find the example app here.

Contributing

Contributions are welcome. In case of any problems look at existing issues, if you cannot find anything related to your problem then open an issue. Create an issue before opening a pull request for non-trivial fixes. In case of trivial fixes open a pull request directly.