Awesome

This is an example how to run a TensorFlow object detection model as web server in TensorFlow C++ API.

Brief Introduction

1). Construct a class for TF model like this:

class Detector {
    std::unique_ptr<tensorflow::Session> session;
    public:
        int loadModel();
        int detect();
};

First, loadModel to initialize and load graph into session, then use session in detect for prediction.

2). use crow the start a web server in main (crow is inspired by python FLASK. If you are familiar with FLASK, crow is easy to use.)

Requirements

• TensorFlow 1.8.0rc1
• OpenCV 3.4.0
• Boost 1.5.8
• Ubuntu 16.04

Install

Install TensorFlow C++ and OpenCV: see this blog

Install Boost <br>

sudo apt-get install libboost-all-dev

Usage

1. compile the project <br>

cmake .
make

2. run tf-cpp web service <br>

./tf_detect_crow

3. test with python script<br>

python test_cpp_api.py

Acknowledgement

Great appreciation to following project and code snippet:

• opencv_tensor.cc
• tensorflow object detection cpp
• crow example
• Encoding and decoding base 64 with c++