• Sample project for Inference Helper (https://github.com/iwatake2222/InferenceHelper )
• Run a simple classification model (MobileNetv2) using several deep leraning frameworks

./main [input]

 - option description: [input]
    - blank
        - use the default image file set in source code (main.cpp)
        - e.g. `./main`
     - *.mp4, *.avi, *.webm
        - use video file
        - e.g. `./main test.mp4`
    - *.jpg, *.png, *.bmp
        - use image file
        - e.g. `./main test.jpg`
    - number (e.g. 0, 1, 2, ...)
        - use camera
        - e.g. `./main 0`

• OpenCV 4.x

• Get source code

  git clone https://github.com/iwatake2222/InferenceHelper_Sample
  cd InferenceHelper_Sample
  git submodule update --init
  sh InferenceHelper/third_party/download_prebuilt_libraries.sh

  • If you have a problem, please refer to https://github.com/iwatake2222/InferenceHelper#installation
  • If your host PC is Windows but you want to build/run on Linux (like WSL2), it's better to run this script on the target OS(Linux). Otherwise, symbolic link will broken.
• Download models

  sh ./download_resource.sh

cd pj_cls_mobilenet_v2
mkdir -p build && cd build
cmake .. -DINFERENCE_HELPER_ENABLE_MNN=on
make
./main

• Configure and Generate a new project using cmake-gui for Visual Studio 2019 64-bit
  • Where is the source code : path-to-InferenceHelper_Sample/pj_cls_mobilenet_v2
  • Where to build the binaries : path-to-build (any)
  • Check one of the liseted InferenceHelperFramework (e.g. INFERENCE_HELPER_ENABLE_MNN )
• Open main.sln
• Set main project as a startup project, then build and run!

sudo apt install g++-arm-linux-gnueabi g++-arm-linux-gnueabihf g++-aarch64-linux-gnu

export CC=aarch64-linux-gnu-gcc
export CXX=aarch64-linux-gnu-g++
cmake .. -DBUILD_SYSTEM=aarch64 -DINFERENCE_HELPER_ENABLE_MNN=on

export CC=arm-linux-gnueabi-gcc
export CXX=arm-linux-gnueabi-g++
cmake .. -DBUILD_SYSTEM=armv7 -DINFERENCE_HELPER_ENABLE_MNN=on

You need to link appropreate OpenCV.

• Requirements

  • Android Studio
    • Compile Sdk Version
      • 30
    • Build Tools version
      • 30.0.0
    • Target SDK Version
      • 30
    • Min SDK Version
      • 24
      • With 23, I got the following error
        • bionic/libc/include/bits/fortify/unistd.h:174: undefined reference to __write_chk'`
        • android/ndk#1179
  • Android NDK
    • 23.1.7779620
  • OpenCV
    • opencv-4.3.0-android-sdk.zip
  • *The version is just the version I used

• Configure NDK

  • File -> Project Structure -> SDK Location -> Android NDK location (before Android Studio 4.0)
    • C:\Users\abc\AppData\Local\Android\Sdk\ndk\21.3.6528147
  • Modify local.properties to specify sdk.dir and ndk.dir (after Android Studio 4.1)

    sdk.dir=C\:\\Users\\xxx\\AppData\\Local\\Android\\Sdk
    ndk.dir=C\:\\Users\\xxx\\AppData\\Local\\Android\\sdk\\ndk\\23.1.7779620
    

• Import OpenCV

  • Download and extract OpenCV android-sdk (https://github.com/opencv/opencv/releases )
  • File -> New -> Import Module
    • path-to-opencv\opencv-4.3.0-android-sdk\OpenCV-android-sdk\sdk
  • FIle -> Project Structure -> Dependencies -> app -> Declared Dependencies -> + -> Module Dependencies
    • select sdk
  • In case you cannot import OpenCV module, remove sdk module and dependency of app to sdk in Project Structure

• Note: To avoid saving modified settings, use the following command

  • git update-index --skip-worktree ViewAndroid/app/build.gradle ViewAndroid/settings.gradle ViewAndroid/.idea/gradle.xml

• Copy resource directory to /storage/emulated/0/Android/data/com.iwatake.viewandroidinferencehelpersample/files/Documents/resource

  • the directory will be created after running the app (so the first run should fail because model files cannot be read)

• Modify ViewAndroid\app\src\main\cpp\CMakeLists.txt to select a image processor you want to use

  • set(ImageProcessor_DIR "${CMAKE_CURRENT_LIST_DIR}/../../../../../pj_cls_mobilenet_v2/image_processor")
  • replace pj_cls_mobilenet_v2 to another

• Choose one of the following options.
  • Note : InferenceHelper itself supports multiple frameworks (i.e. you can set on for several frameworks). However, in this sample project the selected framework is also used to create InferenceHelper instance for the sake of ease.
  • Note : When you change an option, it's safer to clean the project before you re-run cmake

cmake .. -DINFERENCE_HELPER_ENABLE_OPENCV=on
cmake .. -DINFERENCE_HELPER_ENABLE_TFLITE=on
cmake .. -DINFERENCE_HELPER_ENABLE_TFLITE_DELEGATE_XNNPACK=on
cmake .. -DINFERENCE_HELPER_ENABLE_TFLITE_DELEGATE_GPU=on
cmake .. -DINFERENCE_HELPER_ENABLE_TFLITE_DELEGATE_EDGETPU=on
cmake .. -DINFERENCE_HELPER_ENABLE_TFLITE_DELEGATE_NNAPI=on
cmake .. -DINFERENCE_HELPER_ENABLE_TENSORRT=on
cmake .. -DINFERENCE_HELPER_ENABLE_NCNN=on
cmake .. -DINFERENCE_HELPER_ENABLE_MNN=on
cmake .. -DINFERENCE_HELPER_ENABLE_SNPE=on
cmake .. -DINFERENCE_HELPER_ENABLE_ARMNN=on
cmake .. -DINFERENCE_HELPER_ENABLE_NNABLA=on
cmake .. -DINFERENCE_HELPER_ENABLE_NNABLA_CUDA=on
cmake .. -DINFERENCE_HELPER_ENABLE_ONNX_RUNTIME=on
cmake .. -DINFERENCE_HELPER_ENABLE_ONNX_RUNTIME_CUDA=on
cmake .. -DINFERENCE_HELPER_ENABLE_LIBTORCH=on
cmake .. -DINFERENCE_HELPER_ENABLE_LIBTORCH_CUDA=on
cmake .. -DINFERENCE_HELPER_ENABLE_TENSORFLOW=on
cmake .. -DINFERENCE_HELPER_ENABLE_TENSORFLOW_GPU=on

• You may need something like the following commands to run the app

  cp libedgetpu.so.1.0 libedgetpu.so.1
  sudo LD_LIBRARY_PATH=./ ./main

• Build for Android
  • In case you encounter error: use of typeid requires -frtti error, modify ViewAndroid\sdk\native\jni\include\opencv2\opencv_modules.hpp
    • //#define HAVE_OPENCV_FLANN

• InferenceHelper_Sample
• https://github.com/iwatake2222/InferenceHelper_Sample
• Copyright 2020 iwatake2222
• Licensed under the Apache License, Version 2.0

• This project utilizes OSS (Open Source Software)
  • NOTICE.md