tflite_plus 1.0.4

tflite_plus: ^1.0.4 copied to clipboard

Static Analysis & Example App Fixes 🧹 #

flutter analyze Clean-up

• 324 → 0 issues: Resolved every warning/error reported by flutter analyze across the plugin and all 14 bundled example apps
• Plugin lints: Removed the unnecessary named library directive and escaped a stray List<dynamic> doc-comment that was being parsed as HTML
• Deprecated API usage: Replaced Color.withOpacity() with Color.withValues(alpha: ...) across 11 example apps
• Missing type annotations: Added explicit parameter types in the reinforcement_learning example

Example App Fixes

• Broken dependency: object_detection_ssd_mobilenet_v2 pointed at a nonexistent tflite_plus: ^0.10.4 on pub.flutter-io.cn; switched to a local path dependency like every other example
• Version conflict: digit_classification pinned image: 4.0.17, which conflicted with the plugin's own image: ^4.1.7 requirement and blocked pub get entirely
• .gitignore bug: the negation patterns meant to keep placeholder assets/models/.gitkeep and README.md files tracked were missing their ! prefix, so those directories could never actually ship; fixed the patterns and added the missing placeholders

Dependencies

• Bumped ffigen dev dependency to ^21.0.0

Enhanced Documentation & Examples 📚 #

Documentation Improvements

• Comprehensive Examples README: Complete rewrite of example/README.md with detailed documentation
  • ✅ 15+ Example Showcase: Comprehensive catalog of all available ML examples
  • ✅ Feature Matrix: Detailed comparison of capabilities across Computer Vision, Audio, NLP, and Advanced AI
  • ✅ Platform Support: Clear platform compatibility matrix for all examples
  • ✅ Quick Start Guide: Step-by-step setup instructions with code examples
  • ✅ Live Code Samples: Ready-to-use code snippets for:
    • Image Classification with MobileNet
    • Real-time Object Detection with camera streams
    • Audio Classification with microphone input
    • Text Classification and sentiment analysis
    • Proper resource management and error handling

Examples Coverage

• Computer Vision: 10 examples (Image Classification, Object Detection, Pose Estimation, Segmentation, Style Transfer, etc.)
• Audio Processing: YAMNet audio classification with live stream support
• Natural Language Processing: Text classification and BERT Q&A examples
• Advanced AI: Reinforcement learning and gesture recognition demos

Developer Experience

• Setup Instructions: Detailed platform-specific configuration guides
• Troubleshooting Section: Common issues and solutions
• Best Practices: Performance optimization and resource management tips
• Contributing Guidelines: Clear instructions for adding new examples
• Learning Resources: Educational content for ML concepts and implementation patterns

Technical Details

• Cross-Platform Support: Updated compatibility information for Android, iOS, and Desktop platforms
• Hardware Acceleration: Documented GPU, NNAPI, Metal, and CoreML delegate usage
• Model Management: Guidelines for downloading and managing TensorFlow Lite models
• Performance Optimization: Memory usage and inference speed optimization techniques

Major API Overhaul - FFI Implementation 🔄 #

Breaking Changes

• Complete API Rewrite: Migrated from high-level method channel API to low-level FFI-based Interpreter API
• Removed Legacy API: All TflitePlus.* static methods have been removed:
  • ❌ TflitePlus.loadModel()
  • ❌ TflitePlus.runModelOnImage()
  • ❌ TflitePlus.detectObjectOnImage()
  • ❌ TflitePlus.runPoseNetOnImage()
  • ❌ TflitePlus.getAvailableDelegates()
  • ❌ TflitePlus.close()

New Features

• FFI Interpreter API: Direct FFI bindings to TensorFlow Lite C++ library
  • ✅ Interpreter.fromAsset() - Load models from Flutter assets
  • ✅ Interpreter.fromFile() - Load models from file system
  • ✅ Interpreter.fromBuffer() - Load models from memory buffer
  • ✅ interpreter.run() - Single input/output inference
  • ✅ interpreter.runForMultipleInputs() - Multi-input/output inference
  • ✅ interpreter.invoke() - Raw inference execution
• Hardware Delegates: Platform-specific acceleration
  • ✅ GpuDelegate (Android)
  • ✅ MetalDelegate (iOS)
  • ✅ XNNPackDelegate (Cross-platform)
  • ✅ CoreMLDelegate (iOS)
• InterpreterOptions: Configuration for threads, delegates, and optimization
• Direct Tensor Access: Low-level tensor manipulation with Tensor class
• Model Management: Model class for advanced model operations

Migration Guide

// Old API (v1.0.0)
await TflitePlus.loadModel(model: 'model.tflite');
final results = await TflitePlus.runModelOnImage(path: imagePath);

// New API (v1.0.1+)
final interpreter = await Interpreter.fromAsset('model.tflite');
final input = Float32List(inputSize);
final output = List.filled(outputSize, 0.0);
interpreter.run(input, output);
interpreter.close();

Improvements

• Performance: Direct FFI calls eliminate method channel overhead
• Memory Management: Explicit resource control with close() method
• Type Safety: Strongly typed tensor operations
• Lower-level Access: Full control over inference pipeline
• Cross-platform: Unified API across Android and iOS

Developer Experience

• Updated Documentation: All examples updated to new Interpreter API
• Complete Examples: Real-world image classification and batch processing samples
• Migration Support: Clear migration path from legacy API
• Error Handling: Improved exception handling with specific error types

Initial Release 🎉 #

Features

• Image Classification: Complete support for image classification using TensorFlow Lite models
• Object Detection: Comprehensive object detection with bounding boxes and confidence scores
• Pose Estimation: Human pose estimation using PoseNet models
• Semantic Segmentation: Pixel-level image segmentation capabilities
• Multi-Platform Support: Full Android and iOS compatibility
• Hardware Acceleration:
  • Android: GPU delegate, NNAPI delegate
  • iOS: Metal delegate, CoreML delegate
• Flexible Input Methods: Support for both file paths and binary data
• Asynchronous Operations: Non-blocking inference with proper async/await support
• Model Management: Load, close, and query model information
• Comprehensive API: All major TensorFlow Lite operations covered

Platform Support

• Android: API level 21+ with LiteRT 2.0.2
• iOS: iOS 12.0+ with TensorFlow Lite Swift

Dependencies

• Google AI Edge LiteRT 2.0.2
• TensorFlow Lite Task Vision/Text libraries
• GPU acceleration libraries for both platforms

Documentation

• Comprehensive README with examples
• API documentation for all methods
• Performance optimization guides
• Troubleshooting section
• Example app with practical demonstrations

Developer Experience

• Type-safe Dart APIs
• Comprehensive error handling
• Detailed logging and debugging support
• Easy integration with existing Flutter apps
• Extensive customization options