Google's ML Kit Flutter Plugin

A Flutter plugin to use Google's standalone ML Kit for Android and iOS.

Features

Vision

Feature	Android	iOS
Text Recognition	✅	✅
Face Detection	✅	✅
Pose Detection	✅	✅
Selfie Segmentation	yet	yet
Barcode Scanning	✅	✅
Image Labelling	✅	yet
Object Detection and Tracking	✅	yet
Digital Ink Recognition	✅	✅

Natural Language

Feature	Android	iOS
Language Identification	✅	yet
On-Device Translation	✅	yet
Smart Reply	✅	yet
Entity Extraction	✅	yet

Requirements

iOS:

• Minimum iOS Deployment Target: 10.0
• Xcode 12 or newer
• Swift 5
• ML Kit only supports 64-bit architectures (x86_64 and arm64). Check this list to see if your device has the required device capabilities.

Android:

• minSdkVersion: 21
• targetSdkVersion: 29

Usage

Add this plugin as dependency in your pubspec.yaml.

• In your project-level build.gradle file, make sure to include Google's Maven repository in both your buildscript and allprojects sections(for all api's).
• The plugin has been written using bundled api models, this implies models will be bundled along with plugin and there is no need to implement any dependencies on your part and should work out of the box.

1. Create an InputImage

From path:

final inputImage = InputImage.fromFilePath(filePath);

From file:

final inputImage = InputImage.fromFile(file);

From bytes:

final inputImage = InputImage.fromBytes(bytes: bytes, inputImageData: inputImageData);

From CameraImage (if you are using the camera plugin):

final camera; // your camera instance
final WriteBuffer allBytes = WriteBuffer();
for (Plane plane in cameraImage.planes) {
  allBytes.putUint8List(plane.bytes);
}
final bytes = allBytes.done().buffer.asUint8List();

final Size imageSize = Size(cameraImage.width.toDouble(), cameraImage.height.toDouble());

final InputImageRotation imageRotation =
    InputImageRotationMethods.fromRawValue(camera.sensorOrientation) ??
        InputImageRotation.Rotation_0deg;

final InputImageFormat inputImageFormat =
    InputImageFormatMethods.fromRawValue(cameraImage.format.raw) ??
        InputImageFormat.NV21;

final planeData = cameraImage.planes.map(
  (Plane plane) {
    return InputImagePlaneMetadata(
      bytesPerRow: plane.bytesPerRow,
      height: plane.height,
      width: plane.width,
    );
  },
).toList();

final inputImageData = InputImageData(
  size: imageSize,
  imageRotation: imageRotation,
  inputImageFormat: inputImageFormat,
  planeData: planeData,
);

final inputImage = InputImage.fromBytes(bytes: bytes, inputImageData: inputImageData);

2. Create an instance of detector

// vision
final barcodeScanner = GoogleMlKit.vision.barcodeScanner();
final digitalInkRecogniser = GoogleMlKit.vision.digitalInkRecogniser();
final faceDetector = GoogleMlKit.vision.faceDetector();
final imageLabeler = GoogleMlKit.vision.imageLabeler();
final poseDetector = GoogleMlKit.vision.poseDetector();
final textDetector = GoogleMlKit.vision.textDetector();
final objectDetector = GoogleMlKit.vision.objectDetector(CustomObjectDetectorOptions or ObjectDetectorOptions);

// nl
final entityExtractor = GoogleMlKit.nlp.entityExtractor();
final languageIdentifier = GoogleMlKit.nlp.languageIdentifier();
final onDeviceTranslator = GoogleMlKit.nlp.onDeviceTranslator();
final smartReply = GoogleMlKit.nlp.smartReply();

//  Managing models
final translateLanguageModelManager = GoogleMlKit.nlp.translateLanguageModelManager();
final entityModelManager = GoogleMlKit.nlp.entityModelManager();
final remoteModelManager = GoogleMlKit.vision.remoteModelManager();

3. Call the corresponding method

// vision
final List<Barcode> barcodes = await barcodeScanner.processImage(inputImage);
final List<RecognitionCandidate> canditates = await digitalInkRecogniser.readText(points, languageTag);
final List<Face> faces = await faceDetector.processImage(inputImage);
final List<ImageLabel> labels = await imageLabeler.processImage(inputImage);
final List<Pose> poses = await poseDetector.processImage(inputImage);
final RecognisedText recognisedText = await textDetector.processImage(inputImage);
final List<DetectedObject> objects = await objectDetector.processImage(inputImage);

// nl
final List<EntityAnnotation> entities = await entityExtractor.extractEntities(text, filters, locale, timezone);
final bool response = await entityModelManager.downloadModel(modelTag);
final String response = await entityModelManager.isModelDownloaded(modelTag);
final String response = await entityModelManager.deleteModel(modelTag);
final List<String> availableModels = await entityModelManager.getAvailableModels();
final String response = await languageIdentifier.identifyLanguage(text);
final List<IdentifiedLanguage> response = await languageIdentifier.identifyPossibleLanguages(text);
final String response = await onDeviceTranslator.translateText(text);
final bool response = await translateLanguageModelManager.downloadModel(modelTag);
final String response = await translateLanguageModelManager.isModelDownloaded(modelTag);
final String response = await translateLanguageModelManager.deleteModel(modelTag);
final List<String> availableModels = await translateLanguageModelManager.getAvailableModels();
final List<SmartReplySuggestion> suggestions = await smartReply.suggestReplies();
// add conversations for suggestions
smartReply.addConversationForLocalUser(text);
smartReply.addConversationForRemoteUser(text, userID);

4. Extract data from response.

a. Extract barcodes.

for (Barcode barcode in barcodes) {
  final BarcodeType type = barcode.type;
  final Rect boundingBox = barcode.value.boundingBox;
  final String displayValue = barcode.value.displayValue;
  final String rawValue = barcode.value.rawValue;

  // See API reference for complete list of supported types
  switch (type) {
    case BarcodeType.wifi:
      BarcodeWifi barcodeWifi = barcode.value;
      break;
    case BarcodeValueType.url:
      BarcodeUrl barcodeUrl = barcode.value;
      break;
  }
}

b. Extract faces.

for (Face face in faces) {
  final Rect boundingBox = face.boundingBox;

  final double rotY = face.headEulerAngleY; // Head is rotated to the right rotY degrees
  final double rotZ = face.headEulerAngleZ; // Head is tilted sideways rotZ degrees

  // If landmark detection was enabled with FaceDetectorOptions (mouth, ears,
  // eyes, cheeks, and nose available):
  final FaceLandmark leftEar = face.getLandmark(FaceLandmarkType.leftEar);
  if (leftEar != null) {
    final Point<double> leftEarPos = leftEar.position;
  }

  // If classification was enabled with FaceDetectorOptions:
  if (face.smilingProbability != null) {
    final double smileProb = face.smilingProbability;
  }

  // If face tracking was enabled with FaceDetectorOptions:
  if (face.trackingId != null) {
    final int id = face.trackingId;
  }
}

c. Extract labels.

for (ImageLabel label in labels) {
  final String text = label.text;
  final int index = label.index;
  final double confidence = label.confidence;
}

d. Extract text.

String text = recognisedText.text;
for (TextBlock block in recognisedText.blocks) {
  final Rect rect = block.rect;
  final List<Offset> cornerPoints = block.cornerPoints;
  final String text = block.text;
  final List<String> languages = block.recognizedLanguages;

  for (TextLine line in block.lines) {
    // Same getters as TextBlock
    for (TextElement element in line.elements) {
      // Same getters as TextBlock
    }
  }
}

e. Pose detection

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

f. Digital Ink Recognition

for (final candidate in candidates) {
  final text = candidate.text;
  final score = candidate.score;
}

g. Extract Suggestions

//status implications
//1 = Language Not Supported
//2 = Can't determine a reply
//3 = Successfully generated 1-3 replies
int status = result['status'];

List<SmartReplySuggestion> suggestions = result['suggestions'];

h. Extract Objects

for(DetectedObject detectedObject in _objects){
  final rect = detectedObject.getBoundinBox();
  final trackingId = detectedObject.getTrackingId();

  for(Label label in detectedObject.getLabels()){
    print('${label.getText()} ${label.getConfidence()}');
  }
}

5. Release resources with close().

// vision
barcodeScanner.close();
digitalInkRecogniser.close();
faceDetector.close();
imageLabeler.close();
poseDetector.close();
textDetector.close();
objectDetector.close();

// nl
entityExtractor.close();
languageIdentifier.close();
onDeviceTranslator.close();
smartReply.close();

Example app

Look at this example to see the plugin in action.

Migrating from ML Kit for Firebase

When Migrating from ML Kit for Firebase read this guide. For Android details read this. For iOS details read this.

Known issues

Android

To reduce the apk size read more about it in issue #26. Also look at this.

iOS

If you are using this plugin in your app and any other plugin that requires Firebase, there is a known issues you will encounter a dependency error when running pod install. To read more about it go to issue #27.

Contributing

Contributions are welcome. In case of any problems open an issue. Create a issue before opening a pull request for non trivial fixes. In case of trivial fixes open a pull request directly.

License

MIT