sercant/android-segmentation

A library for running semantic segmentation on Android.

Mobile Segmenter

This repository holds an example of how to use Tensorflow Lite to run segmentation on an Android phone. It also serves as a library which you can use in your application for a simple segmentation job.

Adding as a dependency

Add it in your root build.gradle at the end of repositories:

allprojects {
    repositories {
        // ...
        maven { url 'https://jitpack.io' }
    }
}

Step 2. Add the dependency

dependencies {
    implementation 'com.github.sercant:android-segmentation:-SNAPSHOT'
}

How to use

Put the model config file and the tflite model under the assets folder.

An example model_config.json:

{
  "input": {
    "width": 225,
    "height": 225,
    "channelCount": 3
  },
  "output": {
    "width": 225,
    "height": 225,
    "channelCount": 1
  },
  "model": {
    "filePath": "model.tflite",
    "classCount": 19
  }
}

In your code, initialize the configuration by reading the json file or filling the configuration by hand.

val config: Config? = context.assets
    .open("model_config.json")
    .bufferedReader()
    .use {
        val json = it.readText()
        Segmenter.Config.fromJson(json)
    }
// config can return null so be sure to check the result.

// Initialize the image segmentation
val segmenter = Segmenter(context, config)

// ...

// Segment a frame (Bitmap should be in `Bitmap.Config.RGBA_8888` format)
val result: IntArray = segmenter.segment(image)

// ...

// To uninitialize the image segmentation simply call
segmenter.close()

// After this point the class instance shouldn't be used again.