📸 lite_vision_ai

A lightweight, fast, and elegant Flutter library for on-device image classification using TensorFlow Lite.
Designed to be simple, offline-first, and easy to integrate into production apps with minimal boilerplate.

---

✨ Features

• ⚡ 100% Offline: Runs completely on-device with zero internet dependency.
• 🚀 High Performance: Fast image inference powered by tflite_flutter.
• 🎯 Custom Models: Easily load your own .tflite models and .txt labels.
• 📊 Top-N Predictions: Configurable top prediction confidence percentages.
• 🌐 Cross-Platform: Supports Android and iOS.
• 🛠️ Flutter 3 & Image v4+ Compatible: Built-in modern pixel normalization and safe byte buffers.

---

📦 Installation

Add lite_vision_ai to your pubspec.yaml:

dependencies:
  flutter:
    sdk: flutter
  lite_vision_ai: ^1.0.3

Then run:

flutter pub get

⚙️ Setup & Assets

1. Create an assets/models/ folder in your project root and add your .tflite model and .txt labels file.
2. Declare them in your pubspec.yaml:

flutter:
  uses-material-design: true
  assets:
    - assets/models/model.tflite
    - assets/models/labels.txt

💡 Note: Ensure your labels.txt file lists one class label per line.

🚀 Quick Start

Here is a complete example showing how to initialize, load, and classify an image:

import 'dart:io';
import 'package:flutter/material.dart';
import 'package:lite_vision_ai/lite_vision_ai.dart';

void main() async {
  WidgetsFlutterBinding.ensureInitialized();

  // 1. Instantiate the class
  final vision = LiteVisionAI();

  // 2. Load your TFLite model & labels
  await vision.load(
    model: 'assets/models/model.tflite',
    labels: 'assets/models/labels.txt',
  );

  // 3. Classify an image file
  final imageFile = File('path/to/your/image.jpg');
  await vision.classify(image: imageFile, top: 3);

  // 4. Access prediction results
  if (vision.isReady) {
    print('🏷️ Top Label: ${vision.name}');
    print('🎯 Accuracy: ${vision.accuracy.toStringAsFixed(2)}%');
    print('📊 Top Predictions: ${vision.predictions}');
  }
}

📖 API Reference

Member	Type	Description
load({required model, required labels})	Future<void>	Loads .tflite model and .txt labels from app assets.
classify({required image, int top = 3})	Future<void>	Runs image inference and extracts the top N predicted labels.
name	String	Returns the label of the top-1 prediction.
accuracy	double	Returns the confidence percentage (%) of the top-1 prediction.
predictions	Map<String, double>	Returns top-N labels mapped to their confidence percentages.
isReady	bool	Returns true if the model has been loaded successfully.

💡 Example Project

Check out the example/ directory for a fully working Flutter application demonstrating real-time image picking and classification.

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

Libraries

lite_vision_ai