How to build Tensorflow C++ library for Windows

This tutorial help people who are strugling with building Tensorflow C++ library for Windows and use it with OpenCV. I will show you how to build Tensorflow C++ library for Windows from source using bazel and Visual Studio 2019.

TODO

Tut install
Tut build
Tut run example with tensorflow
Tut run example with tensorflow + opencv

Prerequisites

Windows 10
Visual Studio 2019 and Build Tools
Python 3.8
bazel 3.1.0
MSYS2
Tensorflow 2.4.4 (CPU only)

***Note***: You can change the version of Tensorflow and bazel, due to the compatibility of Tensorflow and bazel, you can check the compatibility [here](https://www.tensorflow.org/install/source_windows#tested_build_configurations).

Step 1: Install Python 3.8

Download Python 3.8 from here.
Install Python 3.8.
Remember to check the box "Add Python 3.8 to PATH" when installing Python 3.8.
Test Python 3.8 by openning CMD and type command:
```
python --version
```

Step 2: Install bazel 3.1.0

Download bazel 3.1.0 from here
Change the file name to bazel.exe
Copy the file to C:\bazel
Add C:\bazel to PATH environment variable.
Test bazel by openning CMD and type command:
```
bazel --version
```

Step 3: Install MSYS2

Download MSYS2 from here.
Install MSYS2.
Add C:\msys64\usr\bin to PATH environment variable.
Test MSYS2 by openning CMD and type command:
```
mintty --version
```

Step 4: Install Visual Studio 2019 and Build Tools

Download Visual Studio 2019 from here.
Download Build Tools from here.
Install Visual Studio 2019 and Build Tools.
Remember to check the box Desktop development with C++, MSVC v142-VS2019 C++ build tools and Windows 10 SDK when installing Visual Studio 2019.

Step 5: Build Tensorflow 2.4.4

Open CMD and type command:

git clone https://github.com/tensorflow/tensorflow -b v2.4.4 --depth 1

Open the project and install requirements listed under REQUIRED_PACKAGES in tensorflow\tools\pip_package\setup.py
Some Bug Fixing Before Compiling:

Before we start, we should edit the TensorFlow source code, because you will end up with linking errors (unresolved external symbols LK2019) as shown below if you follow the typical framework building steps.

Session and SessionOptions symbols are not exported So what are these errors?

Dynamic-link library(DLL) can only expose a maximum of 60,000 symbols which is one of DLL limitation. As TensorFlow has more than 60,000 symbols, so we have to manually choose which symbols to be exposed if they are not exposed from TensorFlow by default.

According to the experience, in order to deploy a TensorFlow model, the TensorFlow Session and SessionOptions Classes are required to be exported. The steps include:

Edit tensorflow\core\public\session.h Firstly, include TensorFlow macro header.

>Secondly, add TF_EXPORT in front of any function definition or class definition that caused the missing symbol error. >2. Edit TensorFlow/core/public/session_options.h > >Same as session.h, include the macro header and add TF_EXPORT in front of any function definition or class definition that caused the missing symbol error. > > > >If you are facing any missing symbol during your development, you can follow similar steps to export function or class and rebuild again. > >Now, we are ready to start the compilation.

Build Tensorflow 2.4.4 by openning CMD and type command:

bazel build --config=opt //tensorflow:tensorflow.dll
bazel build --config=opt //tensorflow:tensorflow.lib
bazel build --config=opt //tensorflow:install_headers

Note: If you got stuck while building tensorflow.dll (the terminal hang at compiling... but the timming still running), you can try to restart the computer and build again (without running bazel clean).
After building, you will get tensorflow.dll and tensorflow.lib at tensorflow\bazel-bin\tensorflow and headers at tensorflow\bazel-bin\tensorflow\include.

Step 6: Build Tensorflow C++ Example

Create a new project in Visual Studio 2019.
Choose Console App and C++ as the project type.
Go to Project Properties -> C/C++ -> General -> Additional Include Directories and add relative path to tensorflow\bazel-bin\tensorflow\include folder and tensorflow\bazel-bin\tensorflow\include\src folder in Tensorflow build output to the list.
Go to Project Properties -> Linker -> General -> Additional Library Directories and add tensorflow\bazel-bin\tensorflow to the list.
Go to Project Properties -> Linker -> Input -> Additional Dependencies and add tensorflow.lib to the list.
Go to Build Events -> Post-Build Event -> Command Line and add xcopy /y /d "<tensorflow.dll relative path>" "$(OutDir)"
Note: Remember to change Configuration to All Configuration and Platform to All Platforms

Create a new file main.cpp and copy the code below:

#define NOMINMAX

#include <iostream>
#include "tensorflow/core/framework/graph.pb.h"
#include "tensorflow/core/public/session.h"
#include "tensorflow/core/public/version.h"
int main() {
    std::cout << tf_git_version() << std::endl;
    std::unique_ptr<tensorflow::Session> session(tensorflow::NewSession({}));
    return 0;
}

The #define NOMINMAX to avoid the error min and max are defined in windef.h and winnt.h in Windows SDK.
Build the project and you will get the Console execution at $(OutDir).

Step 7: Add Opencv

Download Opencv from here.
Extract by running the .exe file and copy the folder opencv\build\include to tensorflow\bazel-bin\tensorflow\include.
Go to Project Properties -> C/C++ -> General -> Additional Include Directories and add relative path to tensorflow\bazel-bin\tensorflow\include\opencv2 folder in Tensorflow build output to the list.
Go to Project Properties -> Linker -> General -> Additional Library Directories and add opencv\build\x64\vc16\lib to the list.
Change Configuration to Debug and Platform to x64.
- Go to Project Properties -> Linker -> Input -> Additional Dependencies and add opencv_world480d.lib to the list.
- Go to Build Events -> Post-Build Event -> Command Line and add xcopy /y /d "<opencv_world480d.dll relative path>" "$(OutDir)"
Change Configuration to Release and Platform to x64.
- Go to Project Properties -> Linker -> Input -> Additional Dependencies and add opencv_world480.lib to the list.
- Go to Build Events -> Post-Build Event -> Command Line and add xcopy /y /d "<opencv_world480.dll relative path>" "$(OutDir)"

Edit file main.cpp and copy the code below:

#define NOMINMAX

#include <iostream>
#include "tensorflow/core/framework/graph.pb.h"
#include "tensorflow/core/public/session.h"
#include "tensorflow/core/public/version.h"
/*
int main() {
    
    return 0;
}
*/
#include <opencv2/opencv.hpp>
#include <iostream>

using namespace cv;
using namespace std;

int main(int argc, char** argv)
{
    std::cout << tf_git_version() << std::endl;
    std::unique_ptr<tensorflow::Session> session(tensorflow::NewSession({}));
    // Read the image file
    Mat image = imread("<Your Test Image Path>");
    // Check for failure
    if (image.empty())
    {
        cout << "Image Not Found!!!" << endl;
        cin.get(); //wait for any key press
        return -1;
    }
    // Show our image inside a window.
    imshow("Image Window Name here", image);

    //// Wait for any keystroke in the window
    waitKey(0);
    return 0;
}

Build the project and you will get the Console execution at $(OutDir).

trancongman276/Building-Tensorflow-C---for-Windows