opencv/doc/tutorials/dnn/dnn_android/dnn_android.markdown
2020-06-08 18:46:27 +00:00

3.6 KiB

How to run deep networks on Android device

@prev_tutorial{tutorial_dnn_halide_scheduling} @next_tutorial{tutorial_dnn_yolo}

Introduction

In this tutorial you'll know how to run deep learning networks on Android device using OpenCV deep learning module.

Tutorial was written for the following versions of corresponding software:

  • Android Studio 2.3.3
  • OpenCV 3.3.0+

Requirements

Create an empty Android Studio project

  • Open Android Studio. Start a new project. Let's call it opencv_mobilenet.

  • Keep default target settings.

  • Use "Empty Activity" template. Name activity as MainActivity with a corresponding layout activity_main.

  • Wait until a project was created. Go to Run->Edit Configurations. Choose USB Device as target device for runs. Plug in your device and run the project. It should be installed and launched successfully before we'll go next. @note Read @ref tutorial_android_dev_intro in case of problems.

Add OpenCV dependency

  • Go to File->New->Import module and provide a path to unpacked_OpenCV_package/sdk/java. The name of module detects automatically. Disable all features that Android Studio will suggest you on the next window.

  • Open two files:

    1. AndroidStudioProjects/opencv_mobilenet/app/build.gradle

    2. AndroidStudioProjects/opencv_mobilenet/openCVLibrary330/build.gradle

    Copy both compileSdkVersion and buildToolsVersion from the first file to the second one.

    compileSdkVersion 14 -> compileSdkVersion 26

    buildToolsVersion "25.0.0" -> buildToolsVersion "26.0.1"

  • Make the project. There is no errors should be at this point.

  • Go to File->Project Structure. Add OpenCV module dependency.

  • Install once an appropriate OpenCV manager from unpacked_OpenCV_package/apk to target device. @code adb install OpenCV_3.3.0_Manager_3.30_armeabi-v7a.apk @endcode

  • Congratulations! We're ready now to make a sample using OpenCV.

Make a sample

Our sample will takes pictures from a camera, forwards it into a deep network and receives a set of rectangles, class identifiers and confidence values in [0, 1] range.

  • First of all, we need to add a necessary widget which displays processed frames. Modify app/src/main/res/layout/activity_main.xml: @include android/mobilenet-objdetect/res/layout/activity_main.xml

  • Put downloaded MobileNetSSD_deploy.prototxt and MobileNetSSD_deploy.caffemodel into app/build/intermediates/assets/debug folder.

  • Modify /app/src/main/AndroidManifest.xml to enable full-screen mode, set up a correct screen orientation and allow to use a camera. @include android/mobilenet-objdetect/AndroidManifest.xml

  • Replace content of app/src/main/java/org/opencv/samples/opencv_mobilenet/MainActivity.java: @include android/mobilenet-objdetect/src/org/opencv/samples/opencv_mobilenet/MainActivity.java

  • Launch an application and make a fun!