tflite-react-native

A React Native library for accessing TensorFlow Lite API. Supports Classification, Object Detection, Deeplab and PoseNet on both iOS and Android.

Table of Contents

• Installation
• Usage
  • Image Classification
  • Object Detection
    • SSD MobileNet
    • YOLO
  • Deeplab
  • PoseNet
• Example

Installation

$ npm install tflite-react-native --save

Since version 0.60 of React Native, native modules are autolinked. So you don't need to do react-native link tflite-react-native anymore.

iOS

Native dependencies are now managed with cocoapods. So, Make sure you have cocoapods:

pod --version

If not, install-it:

[sudo] gem install cocoapods

Then, you can install the dependencies:

cd ios
pod install

Android

React native 0.60 uses AndroidX. And the java part of this module is not (yet) converted to AndroidX. So in order to make it work, you should use jetify.

At the root folder of your react-native project:

npm install --save-dev jetifier
npx jetify

Also, any time your dependencies update you have to jetify again. So like it's advised on Jetify website, you could add this in the scripts section of your package.json:

{
  "postinstall": "npx jetify",
  ...
}

Add models to the project

iOS

In XCode, right click on the project folder, click Add Files to "xxx"..., select the model and label files.

Android

1. In Android Studio (1.0 & above), right-click on the app folder and go to New > Folder > Assets Folder. Click Finish to create the assets folder.

2. Place the model and label files at app/src/main/assets.

3. In android/app/build.gradle, add the following setting in android block.

    aaptOptions {
        noCompress 'tflite'
    }

Usage

import Tflite from 'tflite-react-native';

let tflite = new Tflite();

Load model:

tflite.loadModel({
  model: 'models/mobilenet_v1_1.0_224.tflite',// required
  labels: 'models/mobilenet_v1_1.0_224.txt',  // required
  numThreads: 1,                              // defaults to 1  
},
(err, res) => {
  if(err)
    console.log(err);
  else
    console.log(res);
});

Image classification:

tflite.runModelOnImage({
  path: imagePath,  // required
  imageMean: 128.0, // defaults to 127.5
  imageStd: 128.0,  // defaults to 127.5
  numResults: 3,    // defaults to 5
  threshold: 0.05   // defaults to 0.1
},
(err, res) => {
  if(err)
    console.log(err);
  else
    console.log(res);
});

• Output fomart:

{
  index: 0,
  label: "person",
  confidence: 0.629
}

Object detection:

SSD MobileNet

tflite.detectObjectOnImage({
  path: imagePath,
  model: 'SSDMobileNet',
  imageMean: 127.5,
  imageStd: 127.5,
  threshold: 0.3,       // defaults to 0.1
  numResultsPerClass: 2,// defaults to 5
},
(err, res) => {
  if(err)
    console.log(err);
  else
    console.log(res);
});

Tiny YOLOv2

tflite.detectObjectOnImage({
  path: imagePath,
  model: 'YOLO',
  imageMean: 0.0,
  imageStd: 255.0,
  threshold: 0.3,        // defaults to 0.1
  numResultsPerClass: 2, // defaults to 5
  anchors: [...],        // defaults to [0.57273,0.677385,1.87446,2.06253,3.33843,5.47434,7.88282,3.52778,9.77052,9.16828]
  blockSize: 32,         // defaults to 32 
},
(err, res) => {
  if(err)
    console.log(err);
  else
    console.log(res);
});

• Output format:

x, y, w, h are between [0, 1]. You can scale x, w by the width and y, h by the height of the image.

{
  detectedClass: "hot dog",
  confidenceInClass: 0.123,
  rect: {
    x: 0.15,
    y: 0.33,
    w: 0.80,
    h: 0.27
  }
}

Deeplab

tflite.runSegmentationOnImage({
  path: imagePath,
  imageMean: 127.5,      // defaults to 127.5
  imageStd: 127.5,       // defaults to 127.5
  labelColors: [...],    // defaults to https://github.com/shaqian/tflite-react-native/blob/master/index.js#L59
  outputType: "png",     // defaults to "png"
},
(err, res) => {
  if(err)
    console.log(err);
  else
    console.log(res);
});

• Output format:

  The output of Deeplab inference is Uint8List type. Depending on the outputType used, the output is:

  • (if outputType is png) byte array of a png image

  • (otherwise) byte array of r, g, b, a values of the pixels

PoseNet

Model is from StackOverflow thread.

tflite.runPoseNetOnImage({
  path: imagePath,
  imageMean: 127.5,      // defaults to 127.5
  imageStd: 127.5,       // defaults to 127.5
  numResults: 3,         // defaults to 5
  threshold: 0.8,        // defaults to 0.5
  nmsRadius: 20,         // defaults to 20 
},
(err, res) => {
  if(err)
    console.log(err);
  else
    console.log(res);
});

• Output format:

x, y are between [0, 1]. You can scale x by the width and y by the height of the image.

[ // array of poses/persons
  { // pose #1
    score: 0.6324902,
    keypoints: {
      0: {
        x: 0.250,
        y: 0.125,
        part: nose,
        score: 0.9971070
      },
      1: {
        x: 0.230,
        y: 0.105,
        part: leftEye,
        score: 0.9978438
      }
      ......
    }
  },
  { // pose #2
    score: 0.32534285,
    keypoints: {
      0: {
        x: 0.402,
        y: 0.538,
        part: nose,
        score: 0.8798978
      },
      1: {
        x: 0.380,
        y: 0.513,
        part: leftEye,
        score: 0.7090239
      }
      ......
    }
  },
  ......
]

Release resources:

tflite.close();

Example

Refer to the example.

In order to try the example by yourself, you can do:

git clone https://github.com/shaqian/tflite-react-native.git
cd tflite-react-native/example
npm install

For iOS, install the dependencies:

cd ios
pod install

Then open the example.xcworkspace file with Xcode, and click the play button.

For Android, just do:

npx jetify
react-native run-android