Counting 2,772 Big Data & Machine Learning Frameworks, Toolsets, and Examples...
Suggestion? Feedback? Tweet @stkim1

Last Commit
May. 23, 2018
Nov. 7, 2017



Run neural network on your Android phone using the new NNAPI !

Android 8.1 introduces Neural Networks API (NNAPI). It's very exciting to run a model in the "native" way supported by Android System. :)

DNNLirary is a wrapper of NNAPI ("DNNLibrary" is for "daquexian's NNAPI library). It lets you easily make the use of the new NNAPI introduced in Android 8.1. You can convert your caffemodel into daq format by the dnntools and run the model directly.

The demo consists daq model files for LeNet, ResNet-18 and SqueezeNet, and you can get your own model conveniently from pretrained caffemodel.

For how to use this lib directly in your project, check out Usage (it's at the bottom)


This screenshot is ResNet-18

Screenshot image resnet

This screenshot is LeNet

Screenshot camera mnist


Android 8.1 introduces NNAPI. From my experient it is very efficient on my Pixel. For example, it takes caffe-android-lib an average time of 43.42ms to do a convolution with 20 5*5 filters on 224*224 image, but it only takes 15.45ms for NNAPI -- about 1/3 of caffe-android-lib.

What's more, we can believe depthwise convolution, which is useful on mobile devices, is optimized in NNAPI. It takes caffe-android-lib and NNAPI 82.32ms and 16.93ms respectively to do 5 * 5 depthwise conv on 224 * 224 * 20 input.

However, NNAPI is not friendly to normal Android developers. It is not designed to be used by normal developers directly. So I wrapped it into a library.

With DNNLibrary it's extremely easy to deploy your caffe model on Android 8.1+ phone. Here is my code to deploy the ResNet-18 on phone:

ModelWrapper.readFile(getAssets(), "resnet18");

float[] result = ModelWrapper.predict(inputData);

Only four lines! And the model file is got from pretrained caffemodel by the dnntools.

If you use the "raw" NNAPI, the code will increase dramatically. Setting up a LeNet needs 200+ lines. (For the 200+ lines LeNet you can check out the second commit of this repo)


The DNNLibrary has been published on jcenter.


implementation 'me.daquexian:dnnlibrary:0.1.7'

(for Gradle 3.0+),


compile 'me.daquexian:dnnlibrary:0.1.7'

(for Gradle lower than 3.0)

in your app's build.gradle's dependencies section.


Please make sure the Android System on your phone is 8.1+, or you may want to use API 27 emulator. If you want to compile the demo please use Android Studio 3.0+, the latest version of NDK is necessary.

Latest Releases
 Dec. 9 2017
Update for MR2
 Dec. 7 2017
 Nov. 19 2017
The first beta version
 Nov. 12 2017
 Nov. 12 2017