A Powerful Tool for Integrating TensorFlow Lite Models in Flutter

Announcement

Update: 26 April, 2023

This repo is a TensorFlow managed fork of the tflite_flutter_plugin project by the amazing Amish Garg. The goal of this project is to support our Flutter community in creating machine-learning backed apps with the TensorFlow Lite framework.

This project is currently a work-in-progress as we update it to create a working plugin that meets the latest and greatest Flutter and TensorFlow Lite standards. That said, pull requests and contributions are more than welcome and will be reviewed by TensorFlow or Flutter team members. We thank you for your understanding as we make progress on this update.

Feel free to reach out to us by posting in the issues or discussion areas.

Thanks!

• PaulTR

Overview

TensorFlow Lite Flutter plugin provides a flexible and fast solution for accessing TensorFlow Lite interpreter and performing inference. The API is similar to the TFLite Java and Swift APIs. It directly binds to TFLite C API making it efficient (low-latency). Offers acceleration support using NNAPI, GPU delegates on Android, Metal and CoreML delegates on iOS, and XNNPack delegate on Desktop platforms.

Key Features

• Multi-platform Support for Android and iOS
• Flexibility to use any TFLite Model.
• Acceleration using multi-threading.
• Similar structure as TensorFlow Lite Java API.
• Inference speeds close to native Android Apps built using the Java API.
• Run inference in different isolates to prevent jank in UI thread.

(Important) Initial setup : Add dynamic libraries to your app

Android & iOS

Examples and support now support dynamic library downloads! iOS samples can be run with the commands

flutter build ios & flutter install ios from their respective iOS folders.

Android can be run with the commands

flutter build android & flutter install android

while devices are plugged in.

Note: TFLite may not work in the iOS simulator. It's recommended that you test with a physical device.

When creating a release archive (IPA), the symbols are stripped by Xcode, so the command flutter build ipa may throw a Failed to lookup symbol ... symbol not found error. To work around this:

1. In Xcode, go to Target Runner > Build Settings > Strip Style
2. Change from All Symbols to Non-Global Symbols

TFLite Flutter Helper Library

The helper library has been deprecated. New development underway for a replacement at https://github.com/google/flutter-mediapipe. Current timeline is to have wide support by the end of August, 2023.

Import

import 'package:tflite_flutter/tflite_flutter.dart';

Usage instructions

Import the libraries

In the dependency section of pubspec.yaml file, add tflite_flutter: ^0.10.1 (adjust the version accordingly based on the latest release)

Creating the Interpreter

From asset

Place your_model.tflite in assets directory. Make sure to include assets in pubspec.yaml.

final interpreter = await tfl.Interpreter.fromAsset('assets/your_model.tflite');

Refer to the documentation for info on creating interpreter from buffer or file.

Performing inference

For single input and output

Use void run(Object input, Object output).

// For ex: if input tensor shape [1,5] and type is float32
var input = [[1.23, 6.54, 7.81, 3.21, 2.22]];

// if output tensor shape [1,2] and type is float32
var output = List.filled(1*2, 0).reshape([1,2]);

// inference
interpreter.run(input, output);

// print the output
print(output);

For multiple inputs and outputs

Use void runForMultipleInputs(List<Object> inputs, Map<int, Object> outputs).

var input0 = [1.23];  
var input1 = [2.43];  

// input: List<Object>
var inputs = [input0, input1, input0, input1];  

var output0 = List<double>.filled(1, 0);  
var output1 = List<double>.filled(1, 0);

// output: Map<int, Object>
var outputs = {0: output0, 1: output1};

// inference  
interpreter.runForMultipleInputs(inputs, outputs);

// print outputs
print(outputs)

Closing the interpreter

interpreter.close();

Asynchronous Inference with IsolateInterpreter

To utilize asynchronous inference, first create your Interpreter and then wrap it with IsolateInterpreter.

final interpreter = await Interpreter.fromAsset('assets/your_model.tflite');
final isolateInterpreter =
        await IsolateInterpreter.create(address: interpreter.address);

Both run and runForMultipleInputs methods of isolateInterpreter are asynchronous:

await isolateInterpreter.run(input, output);
await isolateInterpreter.runForMultipleInputs(inputs, outputs);

By using IsolateInterpreter, the inference runs in a separate isolate. This ensures that the main isolate, responsible for UI tasks, remains unblocked and responsive.

Use this package as a library

Depend on it

Run this command:

With Flutter:

 $ flutter pub add tflite_flutter

This will add a line like this to your package's pubspec.yaml (and run an implicit flutter pub get):

dependencies:
  tflite_flutter: ^0.10.1

Alternatively, your editor might support flutter pub get. Check the docs for your editor to learn more.

Import it

Now in your Dart code, you can use:

import 'package:tflite_flutter/tflite_flutter.dart';

Download details:

Author:  tensorflow
Source: https://github.com/tensorflow/flutter-tflite

License: Apache-2.0 license

#flutter #dart