Skip to content

Latest commit

 

History

History
200 lines (168 loc) · 6.59 KB

android.md

File metadata and controls

200 lines (168 loc) · 6.59 KB
Raw

Build for Android

MMDeploy provides cross compile for android platform.

Model converter is executed on linux platform, and SDK is executed on android platform.

Here are two steps for android build.

  1. Build model converter on linux, please refer to How to build linux

  2. Build SDK using android toolchain on linux.

This doc is only for how to build SDK using android toolchain on linux.

Build From Source

Install Toolchains

  • cmake

    Make sure cmake version >= 3.14.0. If not, you can follow instructions below to install cmake 3.20.0. For more versions of cmake, please refer to cmake website.

    wget https://github.com/Kitware/CMake/releases/download/v3.20.0/cmake-3.20.0-linux-x86_64.tar.gz
tar -xzvf cmake-3.20.0-linux-x86_64.tar.gz
sudo ln -sf $(pwd)/cmake-3.20.0-linux-x86_64/bin/* /usr/bin/

  • ANDROID NDK 19+

    Make sure android ndk version >= 19.0. If not, you can follow instructions below to install android ndk r23c. For more versions of android ndk, please refer to android ndk website.

    wget https://dl.google.com/android/repository/android-ndk-r23c-linux.zip
unzip android-ndk-r23c-linux.zip
cd android-ndk-r23c
export NDK_PATH=${PWD}

Install Dependencies

Install Dependencies for SDK

You can skip this chapter if only interested in model converter.

NAME INSTALLATION
OpenCV
(>=3.0) 

export OPENCV_VERSION=4.6.0
wget https://github.com/opencv/opencv/releases/download/${OPENCV_VERSION}/opencv-${OPENCV_VERSION}-android-sdk.zip
unzip opencv-${OPENCV_VERSION}-android-sdk.zip
export OPENCV_ANDROID_SDK_DIR=${PWD}/OpenCV-android-sdk
ncnn A high-performance neural network inference computing framework supporting for android.
Now, MMDeploy supports 20220721 and has to use git clone to download it. For supported android ABI, see here . 

git clone -b 20220721 https://github.com/Tencent/ncnn.git
cd ncnn
git submodule update --init
export NCNN_DIR=${PWD}

export ANDROID_ABI=arm64-v8a


mkdir -p build_${ANDROID_ABI}
cd build_${ANDROID_ABI}


cmake -DCMAKE_TOOLCHAIN_FILE=${NDK_PATH}/build/cmake/android.toolchain.cmake -DANDROID_ABI="${ANDROID_ABI}" -DANDROID_PLATFORM=android-30 -DNCNN_VULKAN=ON -DNCNN_DISABLE_EXCEPTION=OFF -DNCNN_DISABLE_RTTI=OFF ..
make -j$(nproc) install

OpenJDK It is necessary for building Java API.
See Java API build for building tutorials.

Build MMDeploy

Build Options Spec

NAME VALUE DEFAULT REMARK
MMDEPLOY_BUILD_SDK {ON, OFF} OFF Switch to build MMDeploy SDK
MMDEPLOY_BUILD_SDK_PYTHON_API {ON, OFF} OFF switch to build MMDeploy SDK python package
MMDEPLOY_BUILD_SDK_JAVA_API {ON, OFF} OFF switch to build MMDeploy SDK Java API
MMDEPLOY_BUILD_TEST {ON, OFF} OFF Switch to build MMDeploy SDK unittest cases
MMDEPLOY_TARGET_DEVICES {"cpu"} cpu Enable target device.
If you want use ncnn vulkan accelerate, you still fill {"cpu"} here. Because, vulkan accelerate is only for ncnn net. The other part of inference is still using cpu.
MMDEPLOY_TARGET_BACKENDS {"ncnn"} N/A Enabling inference engine.
By default, no target inference engine is set, since it highly depends on the use case.
Only ncnn backend is supported for android platform now.
After specifying the inference engine, it's package path has to be passed to cmake as follows,
1. ncnn: ncnn. ncnn_DIR is needed. 
-Dncnn_DIR=${NCNN_DIR}/build_${ANDROID_ABI}/install/lib/cmake/ncnn
MMDEPLOY_CODEBASES {"mmcls", "mmdet", "mmseg", "mmedit", "mmocr", "all"} N/A Enable codebase's postprocess modules. It MUST be set by a semicolon separated list of codebase names. The currently supported codebases are 'mmcls', 'mmdet', 'mmedit', 'mmseg', 'mmocr'. Instead of listing them one by one, you can also pass all to enable them all, i.e., -DMMDEPLOY_CODEBASES=allPlease manually edit csrc/mmdeploy/apis/java/native/CMakeLists.txt to avoid compilation errors.
MMDEPLOY_SHARED_LIBS {ON, OFF} ON Switch to build shared library or static library of MMDeploy SDK. Now you should build static library for android. Bug will be fixed soon.

Build SDK and Demos

MMDeploy provides a recipe as shown below for building SDK with ncnn as inference engine for android.

  • cpu + ncnn 
    export ANDROID_ABI=arm64-v8a
cd ${MMDEPLOY_DIR}
mkdir -p build_${ANDROID_ABI} && cd build_${ANDROID_ABI}
cmake .. \
    -DMMDEPLOY_BUILD_SDK=ON \
    -DMMDEPLOY_BUILD_SDK_JAVA_API=ON \
    -DMMDEPLOY_BUILD_EXAMPLES=ON \
    -DOpenCV_DIR=${OPENCV_ANDROID_SDK_DIR}/sdk/native/jni/abi-${ANDROID_ABI} \
    -Dncnn_DIR=${NCNN_DIR}/build_${ANDROID_ABI}/install/lib/cmake/ncnn \
    -DMMDEPLOY_TARGET_BACKENDS=ncnn \
    -DMMDEPLOY_CODEBASES=all \
    -DMMDEPLOY_SHARED_LIBS=ON \
    -DCMAKE_TOOLCHAIN_FILE=${NDK_PATH}/build/cmake/android.toolchain.cmake \
    -DANDROID_ABI=${ANDROID_ABI} \
    -DANDROID_PLATFORM=android-30 \
    -DANDROID_CPP_FEATURES="rtti exceptions"

make -j$(nproc) && make install