File Takes an Image Mat file as an input and convert it to tensor.

opencv_tensor.cc

/*
  Following file take opencv mat file as an input and run inception model on it
  Created by : Kumar Shubham
  Date : 27-03-2016
*/

//Loading Opencv fIles for processing

#include <opencv2/opencv.hpp>
#include <opencv2/imgproc/imgproc.hpp>

#include <string>
#include <iostream>

#include "tensorflow/core/public/session.h"
#include "tensorflow/core/platform/env.h"
#include "tensorflow/core/framework/tensor.h"
#include <fstream>

int main(int argc, char** argv)
{
  
  // Loading the file path provided in the arg into a mat objects
  std::string path = argv[1];
  cv::Mat readImage = cv::imread(path);
  std::cerr << "read image=" << path << std::endl;

  // converting the image to the necessary dim and for normalization
  int height = 299;
  int width = 299;
  int mean = 128;
  int std = 128;

  cv::Size s(height,width);
  cv::Mat Image;
  std::cerr << "resizing\n";
  cv::resize(readImage,Image,s,0,0,cv::INTER_CUBIC);
  std::cerr << "success resizing\n";

  int depth = Image.channels();

  //std::cerr << "height=" << height << " / width=" << width << " / depth=" << depth << std::endl;
  
  // creating a Tensor for storing the data
  tensorflow::Tensor input_tensor(tensorflow::DT_FLOAT, tensorflow::TensorShape({1,height,width,depth}));
  auto input_tensor_mapped = input_tensor.tensor<float, 4>();
  
  cv::Mat Image2;
  Image.convertTo(Image2, CV_32FC1);
  Image = Image2;
  Image = Image-mean;
  Image = Image/std;
  const float * source_data = (float*) Image.data;
  
  // copying the data into the corresponding tensor
  for (int y = 0; y < height; ++y) {
    const float* source_row = source_data + (y * width * depth);
    for (int x = 0; x < width; ++x) {
      const float* source_pixel = source_row + (x * depth);
      for (int c = 0; c < depth; ++c) {
  const float* source_value = source_pixel + c;
  input_tensor_mapped(0, y, x, c) = *source_value;
      }
    }
  }

  // initializing the graph 
  tensorflow::GraphDef graph_def;

  // Name of the folder in which inception graph is present
  std::string graphFile = "../../model/tensorflow_inception_graph.pb";

  // Loading the graph to the given variable
  tensorflow::Status graphLoadedStatus = ReadBinaryProto(tensorflow::Env::Default(),graphFile,&graph_def);
  if (!graphLoadedStatus.ok()){
    std::cout << graphLoadedStatus.ToString()<<std::endl;
    return 1;
  }

  // creating a session with the grap
  std::unique_ptr<tensorflow::Session> session_inception(tensorflow::NewSession(tensorflow::SessionOptions()));
  //session->reset(tensorflow::NewSession(tensorflow::SessionOptions()));
  tensorflow::Status session_create_status = session_inception->Create(graph_def);
  
  if (!session_create_status.ok()){
    return 1;
  }
  // running the loaded graph 
  std::vector<tensorflow::Tensor> finalOutput;

  std::string InputName = "Mul";
  std::string OutputName = "softmax";
  tensorflow::Status run_status  = session_inception->Run({{InputName,input_tensor}},{OutputName},{},&finalOutput);

  // finding the labels for prediction
  std::cerr << "final output size=" << finalOutput.size() << std::endl;
  tensorflow::Tensor output = std::move(finalOutput.at(0));
  auto scores = output.flat<float>();
  std::cerr << "scores size=" << scores.size() << std::endl;

  // Label File Name
  std::string labelfile = "../../model/imagenet_comp_graph_label_strings.txt";
  std::ifstream label(labelfile);
  std::string line;
  
  // sorting the file to find the top labels
  std::vector<std::pair<float,std::string>> sorted;

  for (unsigned int i =0; i<=1000 ;++i){
    std::getline(label,line);
    sorted.emplace_back(scores(i),line);
    //std::cout << scores(i) << " / line=" << line << std::endl;
  }

  std::sort(sorted.begin(),sorted.end());
  std::reverse(sorted.begin(),sorted.end());
  std::cout << "size of the sorted file is "<<sorted.size()<< std::endl;
  for(unsigned int i =0 ; i< 5;++i){

    std::cout << "The output of the current graph has category  " << sorted[i].second << " with probability "<< sorted[i].first << std::endl; 
  }
  
  /*cv::namedWindow("imageOpencv",CV_WINDOW_KEEPRATIO);
  cv::imshow("imgOpencv",Image);
  
  */
}

I got this working in a different use case after changing line #63 to convert OpenCV's BGR channel order to TensorFlow's RGB.

BTW, did you find any solution to the issue with JPEG files: tensorflow/tensorflow#1924

@ilaripih What did you do to #63? I'm getting segfaults...

thank you for this sample. Could you please provide the BUILD file?

I want to ask a question. when I use opencv2 to read an image and bazel build my cc file, it turned out that "undefied reference to 'cv::....'. It seems all cv-related function cannot be used. How can I compile opencv2 with my tensorflow? can some one help me ?

@XiZiler and @e1s9 , Check this link (http://126kr.com/article/9ckv0k9msfa) for knowing how to integrate OpenCV.

Thank you for sharing this sample. ^^

As @ilaripih mentioned,

copying the data should be re-ordered from BGR to RGB.

  // copying the data into the corresponding tensor
  for (int y = 0; y < height; ++y) {
    const float* source_row = source_data + (y * width * depth);
    for (int x = 0; x < width; ++x) {
      const float* source_pixel = source_row + (x * depth);
      const float* source_B = source_pixel + 0;
      const float* source_G = source_pixel + 1;
      const float* source_R = source_pixel + 2;

      input_tensor_mapped(0, y, x, 0) = *source_R;
      input_tensor_mapped(0, y, x, 1) = *source_G;
      input_tensor_mapped(0, y, x, 2) = *source_B;
    }
  }

@XiZiler: This is a good writeup: https://jacobgil.github.io/deeplearning/tensorflow-cpp

at line 50: Image.convertTo(Image2, CV_32FC1);
why convert to 1 channel?