mirrors/mmdeploy
1
0
Fork 0
mirror of https://github.com/open-mmlab/mmdeploy.git synced 2025-01-14 08:09:43 +08:00
Code Issues Projects Releases Wiki Activity

update document for building android sdk (#817)

Browse Source 
Co-authored-by: dwSun <dwsunny@icloud.com>
This commit is contained in:
孙德伟 2022-07-28 21:11:19 +08:00 committed by GitHub
parent b6b22a1b6f
commit f80c90ed47
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
2 changed files with 71 additions and 48 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

56
docs/en/01-how-to-build/android.md
View File

@ -40,12 +40,12 @@ This doc is only for how to build SDK using android toolchain on linux.
- ANDROID NDK 19+
**Make sure android ndk version >= 19.0**. If not, you can follow instructions below to install android ndk r23b. For more versions of android ndk, please refer to [android ndk website](https://developer.android.com/ndk/downloads).
**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](https://developer.android.com/ndk/downloads).
```bash
wget https://dl.google.com/android/repository/android-ndk-r23b-linux.zip
unzip android-ndk-r23b-linux.zip
cd android-ndk-r23b
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=${PWD}
```
@ -67,7 +67,7 @@ You can skip this chapter if only interested in model converter.
<td>OpenCV<br>(>=3.0) </td>
<td>
<pre><code>
export OPENCV_VERSION=4.5.4
export OPENCV_VERSION=4.6.0
wget https://github.com/opencv/opencv/releases/download/${OPENCV_VERSION}/opencv-${OPENCV_VERSION}-android-sdk.zip
unzip opencv-${OPENCV_VERSION}-android-sdk.zip
export OPENCV_ANDROID_SDK_DIR=${PWD}/OpenCV-android-sdk
@ -78,18 +78,23 @@ export OPENCV_ANDROID_SDK_DIR=${PWD}/OpenCV-android-sdk
<tr>
<td>ncnn </td>
<td>A high-performance neural network inference computing framework supporting for android.</br>
<b> Now, MMDeploy supports v20220216 and has to use <code>git clone</code> to download it.</b><br>
<b> Now, MMDeploy supports 20220721 and has to use <code>git clone</code> to download it. For supported android ABI, see <a href='https://github.com/Tencent/ncnn/releases'> here </a>. </b><br>
<pre><code>
git clone -b 20220216 https://github.com/Tencent/ncnn.git
git clone -b 20220721 https://github.com/Tencent/ncnn.git
cd ncnn
git submodule update --init
export NCNN_DIR=${PWD}
mkdir -p build
cd build
cmake -DCMAKE_TOOLCHAIN_FILE=${NDK_PATH}/build/cmake/android.toolchain.cmake -DANDROID_ABI="arm64-v8a" -DANDROID_PLATFORM=android-30 -DNCNN_VULKAN=ON -DNCNN_DISABLE_EXCEPTION=OFF -DNCNN_DISABLE_RTTI=OFF ..
make install
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$(nproc) install
</code></pre>
</td>
</td>
</tr>
<tr>
<td>OpenJDK </td>
@ -151,14 +156,14 @@ make install
<td>Enabling inference engine. <br><b>By default, no target inference engine is set, since it highly depends on the use case.</b><br> Only ncnn backend is supported for android platform now.<br>
After specifying the inference engine, it's package path has to be passed to cmake as follows, <br>
1. <b>ncnn</b>: ncnn. <code>ncnn_DIR</code> is needed.
<pre><code>-Dncnn_DIR=${NCNN_DIR}/build/install/lib/cmake/ncnn</code></pre>
<pre><code>-Dncnn_DIR=${NCNN_DIR}/build_${ANDROID_ABI}/install/lib/cmake/ncnn</code></pre>
</td>
</tr>
<tr>
<td>MMDEPLOY_CODEBASES</td>
<td>{"mmcls", "mmdet", "mmseg", "mmedit", "mmocr", "all"}</td>
<td>N/A</td>
<td>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 <code>all</code> to enable them all, i.e., <code>-DMMDEPLOY_CODEBASES=all</code></td>
<td>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 <code>all</code> to enable them all, i.e., <code>-DMMDEPLOY_CODEBASES=all</code>Please manually edit <code>csrc/mmdeploy/apis/java/native/CMakeLists.txt</code> to avoid compilation errors.</td>
</tr>
<tr>
<td>MMDEPLOY_SHARED_LIBS</td>
@ -175,18 +180,19 @@ MMDeploy provides a recipe as shown below for building SDK with ncnn as inferenc
- cpu + ncnn
```Bash
export ANDROID_ABI=arm64-v8a
cd ${MMDEPLOY_DIR}
mkdir -p build && cd build
mkdir -p build_${ANDROID_ABI} && cd build_${ANDROID_ABI}
cmake .. \
-DMMDEPLOY_BUILD_SDK=ON \
-DMMDEPLOY_BUILD_SDK_JAVA_API=ON \
-DOpenCV_DIR=${OPENCV_ANDROID_SDK_DIR}/sdk/native/jni/abi-arm64-v8a \
-Dncnn_DIR=${NCNN_DIR}/build/install/lib/cmake/ncnn \
-DOpenCV_DIR=${OPENCV_ANDROID_SDK_DIR}/sdk/native/jni/abi-${ANDROID_ABI} \
-Dncnn_DIR=${NCNN_DIR}/build_${ANDROID_ABI}/install/lib/cmake/ncnn \
-DMMDEPLOY_TARGET_BACKENDS=ncnn \
-DMMDEPLOY_CODEBASES=all \
-DMMDEPLOY_SHARED_LIBS=OFF \
-DMMDEPLOY_SHARED_LIBS=ON \
-DCMAKE_TOOLCHAIN_FILE=${NDK_PATH}/build/cmake/android.toolchain.cmake \
-DANDROID_ABI=arm64-v8a \
-DANDROID_ABI=${ANDROID_ABI} \
-DANDROID_PLATFORM=android-30 \
-DANDROID_CPP_FEATURES="rtti exceptions"
@ -196,14 +202,16 @@ MMDeploy provides a recipe as shown below for building SDK with ncnn as inferenc
#### Build Demo
```Bash
cd ${MMDEPLOY_DIR}/build/install/example
export ANDROID_ABI=arm64-v8a
cd ${MMDEPLOY_DIR}/build_${ANDROID_ABI}/install/example
mkdir -p build && cd build
cmake .. \
-DOpenCV_DIR=${OPENCV_ANDROID_SDK_DIR}/sdk/native/jni/abi-arm64-v8a \
-Dncnn_DIR=${NCNN_DIR}/build/install/lib/cmake/ncnn \
-DMMDeploy_DIR=${MMDEPLOY_DIR}/build/install/lib/cmake/MMDeploy \
-DOpenCV_DIR=${OPENCV_ANDROID_SDK_DIR}/sdk/native/jni/abi-${ANDROID_ABI} \
-Dncnn_DIR=${NCNN_DIR}/build_${ANDROID_ABI}/install/lib/cmake/ncnn \
-DMMDeploy_DIR=${MMDEPLOY_DIR}/build_${ANDROID_ABI}/install/lib/cmake/MMDeploy \
-DCMAKE_TOOLCHAIN_FILE=${NDK_PATH}/build/cmake/android.toolchain.cmake \
-DANDROID_ABI=arm64-v8a \
-DANDROID_ABI=${ANDROID_ABI} \
-DANDROID_PLATFORM=android-30
make -j$(nproc)
```

63
docs/zh_cn/01-how-to-build/android.md
View File

@ -40,12 +40,12 @@ MMDeploy 的交叉编译分为两步:
- ANDROID NDK 19+
**保证 android ndk 的版本 >= 19.0**. 如果不是,可以参考以下命令安装 r23b 版本. 如要获取其他版本,请参考 [这里](https://developer.android.com/ndk/downloads)
**保证 android ndk 的版本 >= 19.0**. 如果不是,可以参考以下命令安装 r23c 版本. 如要获取其他版本,请参考 [这里](https://developer.android.com/ndk/downloads)
```bash
wget https://dl.google.com/android/repository/android-ndk-r23b-linux.zip
unzip android-ndk-r23b-linux.zip
cd android-ndk-r23b
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=${PWD}
```
@ -67,7 +67,7 @@ MMDeploy 的交叉编译分为两步:
<td>OpenCV<br>(>=3.0) </td>
<td>
<pre><code>
export OPENCV_VERSION=4.5.4
export OPENCV_VERSION=4.6.0
wget https://github.com/opencv/opencv/releases/download/${OPENCV_VERSION}/opencv-${OPENCV_VERSION}-android-sdk.zip
unzip opencv-${OPENCV_VERSION}-android-sdk.zip
export OPENCV_ANDROID_SDK_DIR=${PWD}/OpenCV-android-sdk
@ -78,18 +78,24 @@ export OPENCV_ANDROID_SDK_DIR=${PWD}/OpenCV-android-sdk
<tr>
<td>ncnn </td>
<td>ncnn 是支持 android 平台的高效神经网络推理计算框架</br>
<b> 目前, MMDeploy 支持 ncnn 的 20220216 版本, 且必须使用<code>git clone</code> 下载源码的方式安装</b><br>
<b> 目前, MMDeploy 支持 ncnn 的 20220721 版本, 且必须使用<code>git clone</code> 下载源码的方式安装。请到 <a href='https://github.com/Tencent/ncnn/releases'> 这里 </a> 查询 ncnn 支持的 android ABI。</b><br>
<pre><code>
git clone -b 20220216 https://github.com/Tencent/ncnn.git
git clone -b 20220721 https://github.com/Tencent/ncnn.git
cd ncnn
git submodule update --init
export NCNN_DIR=${PWD}
mkdir -p build
cd build
cmake -DCMAKE_TOOLCHAIN_FILE=${NDK_PATH}/build/cmake/android.toolchain.cmake -DANDROID_ABI="arm64-v8a" -DANDROID_PLATFORM=android-30 -DNCNN_VULKAN=ON -DNCNN_DISABLE_EXCEPTION=OFF -DNCNN_DISABLE_RTTI=OFF ..
make install
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$(nproc) install
</code></pre>
</td>
</td>
</tr>
<tr>
<td>OpenJDK </td>
@ -120,6 +126,12 @@ make install
<td>OFF</td>
<td>MMDeploy SDK 编译开关</td>
</tr>
<tr>
<td>MMDEPLOY_BUILD_SDK_PYTHON_API</td>
<td>{ON, OFF}</td>
<td>OFF</td>
<td>MMDeploy SDK Python API的编译开关</td>
</tr>
<tr>
<td>MMDEPLOY_BUILD_SDK_JAVA_API</td>
<td>{ON, OFF}</td>
@ -145,14 +157,14 @@ make install
<td>设置推理后端. <br><b>默认情况下,SDK不设置任何后端, 因为它与应用场景高度相关.</b><br> Android 端目前只支持ncnn一个后端 <br>
构建时,几乎每个后端,都需传入一些路径变量,用来查找依赖包. <br>
1. <b>ncnn</b>: 表示 ncnn. 需要设置<code>ncnn_DIR</code>.
<pre><code>-Dncnn_DIR=${NCNN_DIR}/build/install/lib/cmake/ncnn</code></pre>
<pre><code>-Dncnn_DIR=${NCNN_DIR}/build_${ANDROID_ABI}/install/lib/cmake/ncnn</code></pre>
</td>
</tr>
<tr>
<td>MMDEPLOY_CODEBASES</td>
<td>{"mmcls", "mmdet", "mmseg", "mmedit", "mmocr", "all"}</td>
<td>N/A</td>
<td>用来设置 SDK 后处理组件,加载 OpenMMLab 算法仓库的后处理功能. 已支持的算法仓库有'mmcls','mmdet','mmedit','mmseg'和'mmocr'. 如果选择多个codebase,中间使用分号隔开. 比如, 'mmcls', 'mmdet', 'mmedit', 'mmseg', 'mmocr'. 也可以通过 <code>all</code> 的方式, 加载所有codebase, 即 <code>-DMMDEPLOY_CODEBASES=all.</code></td>
<td>用来设置 SDK 后处理组件,加载 OpenMMLab 算法仓库的后处理功能. 已支持的算法仓库有'mmcls','mmdet','mmedit','mmseg'和'mmocr'. 如果选择多个codebase,中间使用分号隔开. 比如, 'mmcls', 'mmdet', 'mmedit', 'mmseg', 'mmocr'. 也可以通过 <code>all</code> 的方式, 加载所有codebase, 即 <code>-DMMDEPLOY_CODEBASES=all.</code>。请同时手动编辑 <code>csrc/mmdeploy/apis/java/native/CMakeLists.txt</code>以避免编译错误。 </td>
</tr>
<tr>
<td>MMDEPLOY_SHARED_LIBS</td>
@ -169,18 +181,19 @@ make install
- cpu + ncnn
```Bash
export ANDROID_ABI=arm64-v8a
cd ${MMDEPLOY_DIR}
mkdir -p build && cd build
mkdir -p build_${ANDROID_ABI} && cd build_${ANDROID_ABI}
cmake .. \
-DMMDEPLOY_BUILD_SDK=ON \
-DMMDEPLOY_BUILD_SDK_JAVA_API=ON \
-DOpenCV_DIR=${OPENCV_ANDROID_SDK_DIR}/sdk/native/jni/abi-arm64-v8a \
-Dncnn_DIR=${NCNN_DIR}/build/install/lib/cmake/ncnn \
-DOpenCV_DIR=${OPENCV_ANDROID_SDK_DIR}/sdk/native/jni/abi-${ANDROID_ABI} \
-Dncnn_DIR=${NCNN_DIR}/build_${ANDROID_ABI}/install/lib/cmake/ncnn \
-DMMDEPLOY_TARGET_BACKENDS=ncnn \
-DMMDEPLOY_CODEBASES=all \
-DMMDEPLOY_SHARED_LIBS=OFF \
-DMMDEPLOY_SHARED_LIBS=ON \
-DCMAKE_TOOLCHAIN_FILE=${NDK_PATH}/build/cmake/android.toolchain.cmake \
-DANDROID_ABI=arm64-v8a \
-DANDROID_ABI=${ANDROID_ABI} \
-DANDROID_PLATFORM=android-30 \
-DANDROID_CPP_FEATURES="rtti exceptions"
@ -190,14 +203,16 @@ make install
#### 编译 Demo
```Bash
cd ${MMDEPLOY_DIR}/build/install/example
export ANDROID_ABI=arm64-v8a
cd ${MMDEPLOY_DIR}/build_${ANDROID_ABI}/install/example
mkdir -p build && cd build
cmake .. \
-DOpenCV_DIR=${OPENCV_ANDROID_SDK_DIR}/sdk/native/jni/abi-arm64-v8a \
-Dncnn_DIR=${NCNN_DIR}/build/install/lib/cmake/ncnn \
-DMMDeploy_DIR=${MMDEPLOY_DIR}/build/install/lib/cmake/MMDeploy \
-DOpenCV_DIR=${OPENCV_ANDROID_SDK_DIR}/sdk/native/jni/abi-${ANDROID_ABI} \
-Dncnn_DIR=${NCNN_DIR}/build_${ANDROID_ABI}/install/lib/cmake/ncnn \
-DMMDeploy_DIR=${MMDEPLOY_DIR}/build_${ANDROID_ABI}/install/lib/cmake/MMDeploy \
-DCMAKE_TOOLCHAIN_FILE=${NDK_PATH}/build/cmake/android.toolchain.cmake \
-DANDROID_ABI=arm64-v8a \
-DANDROID_ABI=${ANDROID_ABI} \
-DANDROID_PLATFORM=android-30
make -j$(nproc)
```