Ubuntu 16.05LTS 环境下TensorFlow在android上运行

 

使用的机子为：

Inter Core i5-3230M CPU 2.6GHz

双系统安装，非虚拟机 Ubuntu 16.05LTS

python2.7

JDK8

安装tensorflow

参考极客http://wiki.jikexueyuan.com/project/tensorflow-zh/get_started/os_setup.html

参考中文社区http://www.tensorfly.cn/tfdoc/get_started/os_setup.html

安装python-pip和python-dev（python2.7）

sudo apt-get install python-pip python-dev

安装numpy

# For Python 2.7:

$ sudo apt-get install python-numpy swigpython-dev python-wheel

安装tensorflow

         官方提供了 Pip, Docker, Virtualenv, Anaconda 或 源码编译的方法安装 TensorFlow，这里选择pip方式安装。

(这里提供镜像有最新的，用极客装的是tensorflow-0.8.0-cp27)

https://mirror.tuna.tsinghua.edu.cn/tensorflow/linux/cpu/

##python 2.7

#Ubuntu/Linux 64-bit, CPU only, Python 2.7:

$ sudopip install --upgradehttps://storage.googleapis.com/tensorflow/linux/cpu/tensorflow-0.8.0-cp27-none-linux_x86_64.whl

 

# Ubuntu/Linux64-bit, GPU enabled, Python 2.7. Requires CUDA toolkit 7.5 and CuDNN v4.

# Forother versions, see "Install from sources" below.

$ sudopip install --upgradehttps://storage.googleapis.com/tensorflow/linux/gpu/tensorflow-0.8.0-cp27-none-linux_x86_64.whl

 

# Mac OSX, CPU only:

$ sudoeasy_install --upgrade six

$ sudopip install --upgradehttps://storage.googleapis.com/tensorflow/mac/tensorflow-0.8.0-py2-none-any.whl

 

SDK与NDK下载

安装Bazel(用来build安卓项目)

https://bazel.build/versions/master/docs/install.html#ubuntu

(ubuntu安装的网站以及步骤或者用以下方法)

 

## Install JDK 8

   sudo add-apt-repository ppa:webupd8team/java

   sudo apt-get update

sudo apt-getinstall oracle-java8-installer

## Add Bazel distribution URI as a packagesource (one time setup)

echo "deb[arch=amd64] http://storage.googleapis.com/bazel-apt stable jdk1.8" | sudotee /etc/apt/sources.list.d/bazel.list

curlhttps://storage.googleapis.com/bazel-apt/doc/apt-key.pub.gpg | sudo apt-key add–

## Update and install Bazel

sudo apt-getupdate && sudo apt-get install bazel

sudo apt-getupgrade bazel

## Install required packages

sudo apt-getinstall pkg-config zip g++ zlib1g-dev unzip

## Compiling

   git clone https://github.com/bazelbuild/bazel.git

   cd bazel

./compile.sh

下载源码

/tensorflow/examples/android是Camera Demo

git clone --recurse-submodules https://github.com/tensorflow/tensorflow

备注：--recurse-submodules 参数是必须得, 用于获取 TesorFlow 依赖的 protobuf 库.

 

下载训练库、配置WORKSPACE、编译

训练库

https://storage.googleapis.com/download.tensorflow.org/models/inception5h.zip

# 解压到/android/assests

unzip inception5h.zip-d ~/tensorflow/tensorflow/examples/android/assets/

配置WORKSPACE

## 配置/tensorflow/WORKSPACE文件，打开并去掉android_sdk_repository和android_ndk_repository的#注释，替换SDK和NDK的路径，配置后内容(部分)如下

 

## 打开配置文件

   gedit ./tensorflow/WORKSPACE

 

## android_sdk_repository和android_ndk_repository配置内容

# Uncomment and update the paths in theseentries to build the Android demo.

android_sdk_repository(

   name = "androidsdk",

   api_level = 24,

   build_tools_version = "24.0.1",

    #Replace with path to Android SDK on your system(替换的SDK路径)

   path = "/home/ubuntu1/Android/Sdk",

)

android_ndk_repository(

name="androidndk",

#(替换的NDK路径)

   path="/home/ubuntu1/Android/android-ndk-r11c",

api_level=21)

编译

## 进入tensorflow根目录进行编译，编译成功后会输出apk到tensorflow/bazel-bin/tensorflow/eexamples/android/

bazel build//tensorflow/examples/android:tensorflow_demo

 

## 如果想输出详尽的编译错误信息需要增加 --verbose_failures

bazel build//tensorflow/examples/android:tensorflow_demo --verbose_failures

 

## If you get build errors about protocolbuffers, run git submodule update --init and build again

git submodule update –init

结语

## 完成以上步骤后环境就搭建完毕了，再一次进行编译

bazel build//tensorflow/examples/android:tensorflow_demo --verbose_failures

## 执行结果

Target//tensorflow/examples/android:tensorflow_demo up-to-date:

 bazel-bin/tensorflow/examples/android/tensorflow_demo_deploy.jar

 bazel-bin/tensorflow/examples/android/tensorflow_demo_unsigned.apk

 bazel-bin/tensorflow/examples/android/tensorflow_demo.apk

INFO: Elapsed time: 3730.512s, CriticalPath: 3464.62s

生成三个文件