platform_interface library

Where the browser plugin plugs Google's JavaScript runtime into the vision task classes: the backend interfaces and factories, and the decoder its results go through. The Android plugin installs only the GPU name reader.

Applications should import mediapipe_vision.dart instead.

Classes

InteractiveSegmenterBackend

One stateful Interactive Segmenter session on an official platform SDK: an image, then complete stroke histories, in submission order.

VisionResultData

A result as an adapter delivers it.

VisionTaskBackend<R>

A serialized, asynchronous adapter to one task of Google's browser runtime, on a worker.

VisionTaskOverlayBackend

A backend that can draw its results into a browser canvas on its worker.

Constants

packedLandmarkStride → const int

Doubles per packed point.

Properties

faceDetectorBackendFactory ↔ VisionBackendFactory<FaceDetectorResult, FaceDetectorOptions>?

Installed by a platform plugin before the first FaceDetector is created.

getter/setter pair

faceLandmarkerBackendFactory ↔ VisionBackendFactory<FaceLandmarkerResult, FaceLandmarkerOptions>?

Installed by a platform plugin before the first FaceLandmarker is created.

getter/setter pair

gestureRecognizerBackendFactory ↔ VisionBackendFactory<GestureRecognizerResult, GestureRecognizerOptions>?

Installed by a platform plugin before the first GestureRecognizer is created.

getter/setter pair

handLandmarkerBackendFactory ↔ VisionBackendFactory<HandLandmarkerResult, HandLandmarkerOptions>?

Installed by a platform plugin before the first HandLandmarker is created.

getter/setter pair

holisticLandmarkerBackendFactory ↔ VisionBackendFactory<HolisticLandmarkerResult, HolisticLandmarkerOptions>?

Installed by a platform plugin before the first HolisticLandmarker is created.

getter/setter pair

imageClassifierBackendFactory ↔ VisionBackendFactory<ImageClassifierResult, ImageClassifierOptions>?

Installed by a platform plugin before the first ImageClassifier is created.

getter/setter pair

imageEmbedderBackendFactory ↔ VisionBackendFactory<ImageEmbedderResult, ImageEmbedderOptions>?

Installed by a platform plugin before the first ImageEmbedder is created.

getter/setter pair

imageSegmenterBackendFactory ↔ VisionBackendFactory<ImageSegmenterResult, ImageSegmenterOptions>?

Installed by a platform plugin before the first ImageSegmenter is created.

getter/setter pair

interactiveSegmenterBackendFactory ↔ Future<InteractiveSegmenterBackend> Function(InteractiveSegmenterOptions)?

Installed by a platform plugin before the first InteractiveSegmenter is created.

getter/setter pair

objectDetectorBackendFactory ↔ VisionBackendFactory<ObjectDetectorResult, ObjectDetectorOptions>?

Installed by a platform plugin before the first ObjectDetector is created.

getter/setter pair

overlayBackends → Expando<VisionTaskOverlayBackend>

The overlay backend behind a task, when its platform has one. Task classes register theirs when they are created; BrowserOverlay looks it up.

final

poseLandmarkerBackendFactory ↔ VisionBackendFactory<PoseLandmarkerResult, PoseLandmarkerOptions>?

Installed by a platform plugin before the first PoseLandmarker is created.

getter/setter pair

taskPlatformGpuReader ↔ Future<String?> Function()?

Names this device's GPU for TaskPlatform.gpu. An adapter package that can read it sets this when it registers, before the platform is first read; it is called once, with the rest of the platform.

getter/setter pair

Functions

attachMaskBuffers(Map<String, dynamic> result, List<ByteBuffer> buffers) → void

Replaces each mask the browser worker named as [width, height, bytes per value, index] in result with [width, height, values] read from buffers: float32 confidences or uint8 categories.

confidenceMaskFromList(List mask) → ConfidenceMask

Packs one [width, height, Float32List] mask list as a confidence mask.

decodeFaceDetectorResult(VisionResultData data) → FaceDetectorResult

Decodes a Face Detector result. Google's boxes are float pixels; the Dart boxes are whole pixels, truncated as the C API does.

decodeFaceLandmarkerResult(VisionResultData data) → FaceLandmarkerResult

Decodes a Face Landmarker result.

decodeGestureRecognizerResult(VisionResultData data) → GestureRecognizerResult

Decodes a Gesture Recognizer result.

decodeHandLandmarkerResult(VisionResultData data) → HandLandmarkerResult

Decodes a Hand Landmarker result.

decodeHolisticLandmarkerResult(VisionResultData data) → HolisticLandmarkerResult

Decodes a Holistic Landmarker result. Google's browser API lists each part per subject; Holistic reports at most one.

decodeImageClassifierResult(VisionResultData data) → ImageClassifierResult

Decodes an Image Classifier result.

decodeImageEmbedderResult(VisionResultData data) → ImageEmbedderResult

Decodes an Image Embedder result: one vector per model head.

decodeImageSegmenterResult(VisionResultData data) → ImageSegmenterResult

Decodes an Image Segmenter result.

decodeObjectDetectorResult(VisionResultData data) → ObjectDetectorResult

Decodes an Object Detector result, boxes as for faces.

decodePoseLandmarkerResult(VisionResultData data) → PoseLandmarkerResult

Decodes a Pose Landmarker result.

unpackLandmarks<T>(List<double> packed, List<int> counts, LandmarkBuilder<T> point) → List<List<T>>

Splits packed into one landmark list per subject.

Typedefs

LandmarkBuilder<T> = T Function({String? name, double? presence, double? visibility, required double x, required double y, required double z})

A landmark constructor: NormalizedLandmark.new or Landmark.new.

VisionBackendFactory<R, O> = Future<VisionTaskBackend<R>> Function(O options)

Creates a task's backend from its options. Installed by a platform plugin before the first task is created; null where the package runs Google's native runtime itself.