pose_detection_tflite #

A pure Dart/Flutter implementation of Google's MediaPipe pose detection and facial landmark models using TensorFlow Lite. This package provides on-device, multi-person pose detection with minimal dependencies, just TensorFlow Lite and image.

Quick Start #

import 'dart:io';
import 'package:pose_detection_tflite/pose_detection_tflite.dart';

Future main() async {
  // 1. initialize
  final detector = PoseDetector();
  await detector.initialize(
    options: const PoseOptions(
      mode: PoseMode.boxesAndLandmarks,
      landmarkModel: PoseLandmarkModel.heavy,
    ),
  );

  // 2. detect
  final imageBytes = await File('path/to/image.jpg').readAsBytes();
  final results = await detector.detect(imageBytes);

  // 3. access results
  for (final pose in results) {
    final bbox = pose.bboxPx;
    print('Bounding box: (${bbox.left}, ${bbox.top}) → (${bbox.right}, ${bbox.bottom})');

    if (pose.hasLandmarks) {
      for (final lm in pose.landmarks) {
        print('${lm.type}: (${lm.x.toStringAsFixed(1)}, ${lm.y.toStringAsFixed(1)}) vis=${lm.visibility.toStringAsFixed(2)}');
      }
    }
  }

  // 4. clean-up
  await detector.dispose();
}

Refer to the sample code on the pub.flutter-io.cn example tab for a more in-depth example.

Pose Detection Modes #

This package supports two operation modes that determine what data is returned:

Pose Landmark Models #

Choose the model that fits your performance needs: