hand_landmarker 3.0.1

A Flutter plugin for real-time hand landmark detection on Android using Google's MediaPipe Hand Landmarker task and a JNI bridge.

Changelog

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

3.0.1 - 2026-07-05

🛠️ Maintenance

• Updates minimum supported SDK version to Flutter 3.44/Dart 3.12.
• Migrates to built-in Kotlin

3.0.0 - 2026-06-19

💥 Breaking Changes

• Asynchronous Streaming API: The synchronous detect() method has been completely removed from the Dart API.
  • Action Required: Consumers must now use the non-blocking, fire-and-forget processFrame() method to pass camera frames, and listen to the new landmarkStream to receive hand tracking data asynchronously.

⚡ Performance Enhancements

• Zero-Blocking Inference: Transitioned the core MediaPipe architecture from IMAGE mode to LIVE_STREAM mode. Inference now runs entirely on a native background worker thread, completely eliminating Flutter UI jank and dropped frames.
• Direct YUV to ARGB Conversion: Replaced the expensive YUV -> NV21 -> JPEG -> Bitmap conversion pipeline with a direct YUV to ARGB8888 Bitmap conversion using fast integer math. This drastically reduces CPU load and memory allocations on every camera frame.
• Optimized Native Bridging: Implemented a Kotlin companion object to capture background results and route them directly to a standard Flutter EventChannel, bypassing complex JNI callback mapping and keeping the bridge lightweight.

2.3.0 - 2026-06-17

🛠️ Maintenance & Dependencies

• JNI Configuration Migration: Migrated the jnigen configuration from jnigen.yaml to a Dart-based script at tool/jnigen.dart for compatibility with jnigen 0.16.0.
• Context Refactor: Refactored native context retrieval in hand_landmarker.dart to use androidApplicationContext via the new jni_flutter package.
• SDK Constraints: Updated the minimum Dart SDK requirement to >=3.3.0 for both the plugin and the example app.
• Dependency Bumps:
  • Bumped jni to ^1.0.0.
  • Added jni_flutter at ^1.0.1.
  • Bumped jnigen (dev dependency) to ^0.16.0.
  • Bumped camera to ^0.12.0+1.
  • Bumped Android MediaPipe tasks-vision dependency to 0.10.29 in android/build.gradle.

2.2.0 - 2025-12-13

💥 Breaking Changes

• Delegate Enum Case Change: The values for the HandLandmarkerDelegate enum have been changed from ALL_CAPS (CPU, GPU) to lowerCamelCase (cpu, gpu) to adhere to Dart linter standards.
  • Action Required: Users must update all usage from HandLandmarkerDelegate.CPU or HandLandmarkerDelegate.GPU to HandLandmarkerDelegate.cpu or HandLandmarkerDelegate.gpu respectively.

🛠️ Maintenance

• Major Build Toolchain Upgrade: Modernized the Android build configuration for both the plugin and the example app. Note: This update requires a compatible development environment, including JDK 17 and Android Studio Giraffe (or newer), to successfully build the plugin.
  • Upgraded Android Gradle Plugin (AGP) to 8.11.1.
  • Retained Kotlin at 2.1.0 due to current jnigen toolchain limitations.
  • Updated Java source/target compatibility and jvmTarget to VERSION_17 (previously VERSION_11).
  • Updated compileSdk to 36.
  • Updated the Gradle Wrapper to version 8.14.
• Dependency Updates: Updated core dependencies: jnito 0.15.2 and jnigen to 0.15.0, and plugin_platform_interface to 2.1.8.

2.1.2 - 2025-11-03

🐛 Bug Fixes

• Fixed potential crashes in Android release builds by adding consumer ProGuard rules. This ensures MediaPipe's essential classes are not stripped by R8. (46a978a)

🛠️ Maintenance

• Updated camera plugin dependency to version 0.11.3.

📝 Documentation

• Added additional examples for plugin usage in README.md.

2.1.1 - 2025-10-27

🐛 Bug Fixes

• Fixed a java.lang.UnsatisfiedLinkError crash on 32-bit (armeabi-v7a) devices by updating the native MediaPipe tasks-vision dependency to 0.10.26.1. (Fixes #1)

2.1.0 - 2025-07-15

✨ Features

• Configurable Options: Added the ability to configure the hand landmarker with the following options:
  • numHands: The maximum number of hands to detect.
  • minHandDetectionConfidence: The minimum confidence score for hand detection to be considered successful.
  • delegate: The delegate to use for inference, allowing for selection between CPU and GPU.
  • see e1dff3e

2.0.0 - 2025-07-08

💥 Breaking Changes

• Synchronous API: The plugin's core methods are now synchronous to improve performance. This affects how you create, use, and dispose of the plugin.
  • HandLandmarkerPlugin.create() no longer returns a Future.
  • HandLandmarkerPlugin.dispose() is now synchronous.
  • The detect() method is now a synchronous, blocking call. You must manage how frequently you call it to avoid blocking the UI thread.

✨ Features & Performance

• Native Image Processing (BREAKING): Rearchitected the plugin to perform all YUV image conversion natively in Kotlin. This eliminates the Dart background isolate and significantly reduces data transfer overhead for much lower latency. (347f5f1)
• GPU Acceleration: Enabled the MediaPipe GPU delegate by default to accelerate model inference, resulting in smoother real-time performance. (4749d8c)

1.0.0

• Initial release of the project.