face_plugin 0.0.2

Face detection and feature extraction plugin using MobileFaceNet for Android and iOS. Provides 128-dimensional face feature vectors for face recognition and comparison.

Changelog

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

0.0.2 - 2026-04-24

Added

• Face.landmarkCount — number of landmarks actually detected by ML Kit (0–5). Use as a quality gate to filter false-positive detections.
• Face.headEulerAngleX/Y/Z — head pose angles from ML Kit (pitch / yaw / roll, degrees). Useful for rejecting non-frontal faces before feature extraction.

Changed

• Face.faceScore is now landmarkCount / 5.0 (real quality signal) instead of a hardcoded 1.0.
• Face.faceTv returns -1 (not 0) when ML Kit tracking is unavailable, avoiding confusion with a real trackingId = 0.
• Android: FEATURE_DIM changed from a static field to an instance field so each plugin instance independently auto-detects the model output shape on attach. Fixes a shape mismatch crash ([1,192] vs [1,128]) on models with 192-dim output.
• Android: detectFaces and extractFeatures now decode the image and run ML Kit on a background ExecutorService thread, with results posted back to the main thread. Eliminates potential "Must not be called on the main application thread" errors and keeps large-image decoding off the UI thread.
• Cleaned up documentation: removed all intermediate dev/debug markdown files. The repository now contains only README.md and CHANGELOG.md.

Fixed

• Unit tests (test/) rewrote to cover the actual detectFaces / extractFeatures API instead of the removed getPlatformVersion() stub.
• Integration test (example/integration_test/) updated accordingly.

---

0.0.1 - 2026-04-07

Added

• Initial release of face_plugin
• Face detection with bounding boxes and facial landmarks (5 key points)
• Feature extraction using MobileFaceNet (128-dimensional vectors)
• Complete Android implementation (Java + TensorFlow Lite 2.14.0)
• Complete iOS implementation (Swift + TensorFlow Lite 2.14.0)
• Cross-platform API through Method Channel
• Comprehensive documentation (README, QUICK_START, MODEL_GUIDE, PROJECT_SUMMARY)
• Example application with UI demonstration
• Support for image preprocessing (resize to 112x112, normalization)

Features

• detectFaces(Uint8List imageBytes) - Detect faces in images
• extractFeatures(Uint8List imageBytes) - Extract 128-D feature vectors
• Face model with bounding boxes, 5 facial landmarks, and confidence scores

Platform Support

• Android: minSdk 21 (Android 5.0+)
• iOS: 12.0+
• Flutter: >=3.3.0

Dependencies

• TensorFlow Lite 2.14.0 for both platforms
• plugin_platform_interface ^2.0.2

Notes

• Current implementation uses simplified face detection (for demonstration)
• For production, integrate professional face detection (ML Kit, MTCNN, Vision Framework)
• Feature extraction is production-ready using MobileFaceNet
• Requires mobilefacenet.tflite model file (not included, see MODEL_GUIDE.md)