face_gesture_detector 0.2.3

A widget that detects face gestures.

Changelog

All notable changes to this project will be documented in this file.

The format loosely follows Keep a Changelog and the project adheres to Semantic Versioning.

0.2.3

Fixed

• Release APK crash (R8/ProGuard): R8 class-merging optimization broke MediaPipe's JNI native library loading via stack inspection, causing ExceptionInInitializerError at runtime. Added -dontoptimize and explicit -keep rules for MediaPipe, protobuf, and plugin classes in consumer ProGuard rules.
• Euler angles rotated 90°: Camera sensor orientation (sensorOrientation=270 on most front cameras) was sent from Dart but never applied on the native side. MediaPipe computed angles in raw sensor coordinates, causing pitch/yaw/roll to be misassigned (e.g. "head nod up" fired when turning right). Now compensated via correctAnglesForRotation() after Euler extraction.
• Distance "ok" range too narrow: Default thresholds (minDistanceRatio=0.20, maxDistanceRatio=0.60) required holding the device uncomfortably close. Relaxed to 0.05/0.40 for comfortable arm-length usage.

---

0.2.2

Fixed

• Corrected CHANGELOG to accurately reflect the 0.2.0 and 0.2.1 release history

---

0.2.1

Fixed

• Deep-cast nested platform channel maps: FaceFrame.fromMap now uses Map<String, dynamic>.from() for nested maps (faceBoundingBox, poseAngles, quality, blendshapes, landmarks) instead of direct as Map<String, dynamic> casts that failed at runtime with _Map<Object?, Object?>
• Face confidence always 0.00: The _neutral blendshape score was incorrectly used as face detection confidence; now reports 1.0 when a face is detected (MediaPipe already filters by minFaceDetectionConfidence)

---

0.2.0

Added

• Camera integration: FaceGestureDetector accepts an optional cameraController parameter, wiring the full camera → native → recognizer pipeline automatically
• Real-time detection: RawFaceGestureDetector manages startDetection, processFrame, faceFrameStream, and stopDetection lifecycle when a CameraController is provided
• Example app: Fully functional demo with live front-camera preview, face detection indicator, event log, and all 10 gesture callbacks
• Integration tests: Camera pipeline tests on physical device (camera init, processFrame, faceFrameStream)

Changed

• cameraController parameter added to FaceGestureDetector and RawFaceGestureDetector (optional — null preserves test/manual mode)
• camera: ^0.11.1 added as a dependency
• Example app requires android.permission.CAMERA

---

0.1.0-dev

Added

• Data Model: FaceFrame, FaceBlendshape (52 values), PoseAngles, FaceLandmark, ImageQualityMetrics
• Details objects: FaceDetectedDetails, QualityDetails, DistanceDetails, PoseDetails, BlinkDetails, SmileDetails, MouthDetails, BrowDetails, HeadTurnDetails, HeadNodDetails
• Configuration: FaceGestureConfiguration with calibrated defaults
• Platform interface: FaceGestureDetectorPlatform + MethodChannelFaceGestureDetector + processFrame()
• Recognizers: FacePresenceRecognizer, QualityGateRecognizer, PoseRecognizer, BlinkRecognizer, SmileRecognizer, MouthRecognizer, BrowRecognizer, HeadTurnRecognizer, HeadNodRecognizer, RawFrameRecognizer
• Widgets: RawFaceGestureDetector (Layer 2), FaceGestureDetector (Layer 1 facade)
• Controller: FaceGestureDetectorController with pause/resume/reset
• Native Android: FaceLandmarkerEngine (MediaPipe LIVE_STREAM + GPU delegate), EulerAngleCalculator (pitch/yaw/roll from 4×4 matrix), FrameDecoder (NV21 → MPImage), SingleSlotFrameBuffer (lock-free backpressure), FaceFrameStreamHandler (EventChannel main-thread emission), face_landmarker.task model bundled
• Example app: MVP with all callback families and event log
• Tests: 117 Dart unit tests, 14 Android JVM unit tests, 3 integration tests on device