Android Studio调用python运行thensorflow模型--CLE方案实现

Android Studio调用python运行thensorflow模型--CLE方案实现

我使用的是虚拟android设备，故对应的CLE中库文件版本为/x86，你可以根据自己开发环境找到对应的版本。调用的python版本为3.7，以下为主要步骤：

1、环境准备

在官网下载最新的CLE for Android 开发包，其中包含示例工程和API文档。

解压开发包，拷贝如下文件

拷贝 starcore_android_rX.X.jar到工程的libs目录。(该jar文件在download的starcore_for_android.3.2.0中)

拷贝libstar_java.so、libstarcore.so、libpython3.7m.so和libstar_python37.so到libs/x86目录。(除了libpython3.7m.so在android.python.3.7.0下载文件中，其余都在download的starcore_for_android.3.2.0/libs/x86中)

2、编写Python代码

/----test.py----/

def add(x,y) :

return x+y

/----calljava.py----/

import imp #test load path

def log(content):

JavaClass.d(“formPython”,content)

log(“Hello Android,form python”)

/----py_code.py----/

import time

def get_time():

return time.time()

将py_code.py压缩为py_code.zip文件。将编写的Python源码放入Android 工程的assets目录，其中还要包含一些Python需要的环境及标准库，见下图：(这些库文件都在android.python.3.7.0下载文件/x86目录下)

如果没有发现assets目录，则如下选择Android下添加：

3、编写Android相关代码，初始化CLE并调用Python

首先需要修改build.gradle文件：

添加jni库文件路径：

// jni配置

sourceSets {

main {

jniLibs.srcDirs = ['libs']

}

}

然后编写Android相关代码，初始化CLE并调用Python

import android.os.Bundle;

import android.support.design.widget.FloatingActionButton;

import android.support.design.widget.Snackbar;

import android.support.v7.app.AppCompatActivity;

import android.support.v7.widget.Toolbar;

import android.view.View;

import android.view.Menu;

import android.view.MenuItem;

import android.os.AsyncTask;

import android.app.Activity;

import android.util.Log;

import java.io.BufferedInputStream;

import java.io.BufferedOutputStream;

import java.io.IOException;

import java.io.InputStream;

import java.io.File;

import java.io.FileOutputStream;

import java.util.*;

import com.srplab.www.starcore.*;

public class MainActivity extends AppCompatActivity {

public StarSrvGroupClass SrvGroup;

@Override

protected void onCreate(Bundle savedInstanceState) {

super.onCreate(savedInstanceState);

setContentView(R.layout.activity_main);

Toolbar toolbar = (Toolbar) findViewById(R.id.toolbar);

setSupportActionBar(toolbar);

FloatingActionButton fab = (FloatingActionButton) findViewById(R.id.fab);

fab.setOnClickListener(new View.OnClickListener() {

@Override

public void onClick(View view) {

Snackbar.make(view, "Replace with your own action", Snackbar.LENGTH_LONG)

.setAction("Action", null).show();

}

});

final File appFile = getFilesDir(); /*-- /data/data/packageName/files --*/

final String appLib = getApplicationInfo().nativeLibraryDir;

loadPy(appFile,appLib);

// AsyncTask.execute(new Runnable() {

//

// @Override

// public void run() {

// loadPy(appFile,appLib);

// }

// });

}

void loadPy(File appFile,String appLib){

//拷贝Python相关环境

File pythonLibFile = new File(appFile, "python3.7.zip");

if (!pythonLibFile.exists()) {

copyFile(this, "python3.7.zip");

copyFile(this, "_struct.cpython-37m.so");

copyFile(this, "binascii.cpython-37m.so");

copyFile(this, "time.cpython-37m.so");

copyFile(this, "zlib.cpython-37m.so");

}

// 拷贝Python 代码

copyFile(this, "calljava.py");

copyFile(this, "test.py");

try {

// 加载Python解释器

System.load(appLib + File.separator + "libpython3.7m.so");

// 除了将代码直接拷贝，还支持将代码压缩为zip包，通过Install方法解压到指定路径

InputStream dataSource = getAssets().open("py_code.zip");

StarCoreFactoryPath.Install(dataSource, appFile.getPath(),true );

} catch (Exception e) {

e.printStackTrace();

}

/*----init starcore----*/

StarCoreFactoryPath.StarCoreCoreLibraryPath = appLib;

StarCoreFactoryPath.StarCoreShareLibraryPath = appLib;

StarCoreFactoryPath.StarCoreOperationPath = appFile.getPath();

StarCoreFactory starcore = StarCoreFactory.GetFactory();

StarServiceClass Service = starcore._InitSimple("test", "123", 0, 0);

SrvGroup = (StarSrvGroupClass) Service._Get("_ServiceGroup");

Service._CheckPassword(false);

/*----run python code----*/

SrvGroup._InitRaw("python37", Service); // this place

StarObjectClass python = Service._ImportRawContext("python", "", false, "");

/*调用tensorflow*/

//个人推荐使用这种写法，看起来比较直观，不容易带给自己误解

python._Call("eval", "import sys");

// python._Call("eval", "sys.path.append(r'" + path + "')");

python._Call("eval", "import numpy as np");

python._Call("eval", "import pred as pd");

//这里在导入tensorflow模型后获得模块接口，可以使用model._Call("func")

//相当于python._Call("eval", "pd.func()"),但是前一种方法没法保存返回结果

StarObjectClass model = python._GetObject("pd");

python._Call("eval", "import operate_data as processor");

// processor = python._GetObject("processor");

// 设置Python模块加载路径

python._Call("import", "sys");

StarObjectClass pythonSys = python._GetObject("sys");

StarObjectClass pythonPath = (StarObjectClass) pythonSys._Get("path");

pythonPath._Call("insert", 0, appFile.getPath()+ File.separator +"python3.7.zip");

pythonPath._Call("insert", 0, appLib);

pythonPath._Call("insert", 0, appFile.getPath());

//调用Python代码

Service._DoFile("python", appFile.getPath() + "/py_code.py", "");

long time = python._Calllong("get_time");

Log.d("", "form python time="+time);

Service._DoFile("python", appFile.getPath() + "/test.py", "");

int result = python._Callint("add", 5, 2);

Log.d("", "result="+result);

python._Set("JavaClass", Log.class);

Service._DoFile("python", appFile.getPath() + "/calljava.py", "");

}

private void copyFile(Activity c, String Name) {

File outfile = new File(c.getFilesDir(), Name);

BufferedOutputStream outStream = null;

BufferedInputStream inStream = null;

try {

outStream = new BufferedOutputStream(new FileOutputStream(outfile));

inStream = new BufferedInputStream(c.getAssets().open(Name));

byte[] buffer = new byte[1024 * 10];

int readLen = 0;

while ((readLen = inStream.read(buffer)) != -1) {

outStream.write(buffer, 0, readLen);

}

} catch (IOException e) {

e.printStackTrace();

} finally {

try {

if (inStream != null) inStream.close();

if (outStream != null) outStream.close();

} catch (IOException e) {

e.printStackTrace();

}

}

}

@Override

public boolean onCreateOptionsMenu(Menu menu) {

// Inflate the menu; this adds items to the action bar if it is present.

getMenuInflater().inflate(R.menu.menu_main, menu);

return true;

}

@Override

public boolean onOptionsItemSelected(MenuItem item) {

// Handle action bar item clicks here. The action bar will

// automatically handle clicks on the Home/Up button, so long

// as you specify a parent activity in AndroidManifest.xml.

int id = item.getItemId();

//noinspection SimplifiableIfStatement

if (id == R.id.action_settings) {

return true;

}

return super.onOptionsItemSelected(item);

}

}

以上是我新建了一个basic activity工程，里面的MainActivity.java文件中初始化python模块并调用文件和函数代码。

代码中python._Set(“JavaClass”, Log.class)一句，指将一个java类设置给Python，变成Python类，第一个参数指定在Python中的类名，此处随意起了一个类名JavaClass，该类可直接在Python中使用，如上将Android的日志输出类设置给Python使用，完成了Java与Python的互相调用。具体tensorflow模型的使用还只是初尝，等我全部走完美了再来补充。

Android Studio调用python运行thensorflow模型--CLE方案实现相关教程