Tensor/IO TensorFlow Android

This library vends a full build of TensorFlow 2.0, v2.0.3 for Android in a Java wrapper. Refer to the r2.0.doc.ai branch of our TensorFlow fork and specifically the Android Build Readme for more info.

Requirements:

Ensure the following have been installed in Android Studio:

• Android NDK 21.1.6352462
• Android Cmake 3.10.2

You can do this in Preferences: Appearance and Behavoir > System Settings > Android SDK > SDK Tools. Click on Show Package Contents at the bottom right and expand the NDK and CMake options to select the appropriate packages.

Additionally the library has a minSdk requirement of API 22.

Installation

If you are downloading this repository over git, a number of large binary files are included with it. You must install Git Large File Storage first and configure it. You can also find the location of the latest copies of the binaries in the jitpack.sh script.

Running the Example Application

Open the project folder in Android Studio, let gradle and all that jazz do its job, select the app configuration, which is probably selected by default, select an x86 or x86_64 emulator or an Android device, and press play.

Using the Library

Jitpack

The simplest way to use this library is to import it into your project as a gradle dependency via jitpack. For detailed instructions see https://jitpack.io/#doc-ai/tensorio-tensorflow-android/8.0

Add the following to your project gradle file:

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

Then add the dependency to your module's gradle file:

dependencies {
  implementation 'com.github.doc-ai:tensorio-tensorflow-android:0.8.0'
  ...
}

And resync your gradle file.

Building the AAR

If you'd prefer not to use Jitpack you can build the AAR yourself and import it directly into a project. We are following these instructions:

https://developer.android.com/studio/projects/android-library

After cloning the repo open up the main project and build and run the app. This will prouduce the following two files. You may need to set your Build Variants appropriately.

tensorflow/build/outputs/aar/tensorflow-release.aar
tensorflow/build/outputs/aar/tensorflow-debug.aar

Next, from the project you want to use this library in, add the AAR. Choose File > New > New Module and select Import .JAR/.AAR Package. Select the tensorflow-release.aar build output from the previous step.

Make sure the library is listed in your settings.gradle file. Android Studio probably takes care of this automatically:

include ':tensorflow-release'
include ':app'

Finally add the library as a dependency in your app's build.gradle file:

dependencies {
	implementation project(":tensorflow-release")
}

You may also need to make the following changes to your app's build.gradle. Ensure the minSdk version is at least API 22:

defaultConfig {
	minSdkVersion 22
}

If you see an error about "More than one file found with OS independent path" add the following to the android section of the gradle file:

packagingOptions {
    pickFirst 'lib/armeabi-v7a/libc++_shared.so'
    pickFirst 'lib/arm64-v8a/libc++_shared.so'
    pickFirst 'lib/x86/libc++_shared.so'
    pickFirst 'lib/x86_64/libc++_shared.so'
}

Running a Model

Additional instructions forthcoming. In the meantime see the included tests and the Java example application.