pose_detection_tflite 1.2.2

pose_detection_tflite: ^1.2.2 copied to clipboard

• Fix Android build: bump tflite_flutter_custom to 1.2.5 (fixes undefined symbol TfLiteIntArrayCreate linker error)
• Add missing MainActivity.kt for Android example app

• Improved auto .dylib bundling on MacOS

• Add native OpenCV preprocessing for 5-15x faster image processing (SIMD-accelerated)
• New useNativePreprocessing parameter on PoseDetector (default: true)
• New detectOnMat() method on YoloV8PersonDetector for direct cv.Mat input
• New runOnMat() method on PoseLandmarkModelRunner for direct cv.Mat input
• Add opencv_dart dependency for native image operations
• Refactored detection code with shared post-processing helpers

• Cache yolo input buffer and pass flat tensor to avoid reshape overhead
• Update tflite_flutter_custom to 1.1.0
• 100% test coverage

• Parallel landmark detection in multi person photos, faster processing
• Improve error handling
• Add integration tests

• Improved Dartdoc coverage

This version contains breaking changes:

• Breaking API changes: PoseDetector uses constructor params (no PoseOptions), returns Pose/BoundingBox, and detection methods are async
• YOLO and BlazePose run via isolates to avoid blocking
• Expanded DartDoc coverage
• poseLandmarkConnections for drawing skeletons.
• Example app improvements, sample images added.

• Removed unused dependency path_provider

• Improved documentation

• Initial release
• Person detection using YOLOv8
• Pose landmark detection with MediaPipe Pose
• Support for lite, full, and heavy models
• Box-only and full landmark detection modes