Deep learning Library to work easily with opencv on window environment.

I was wrapping caffe library to use easy with opencv and window os environment(named MareDeepDLL).

MareDeepDll is referenced on

The dll is made on this environment
Window 10 64bit,
VS 2013 64bit
OpenCV 3.0 64bit
cuda 6.5 64bit
tbb 64bit
All dependency is as follows:

This code is example to use the dll
#include < iostream>  
#include < stdio.h>
#include < vector>
#include < time.h>

#include <  opencv2\opencv.hpp>    
#include <  opencv2\core.hpp>  
#include <  opencv2\highgui.hpp>   
#include <  opencv2\videoio.hpp>    
#include <  opencv2\imgproc.hpp>  

#include "DeepDll_B.h"

#ifdef _DEBUG            
#pragma comment(lib, "opencv_core300d.lib")  
#pragma comment(lib, "opencv_highgui300d.lib")    
#pragma comment(lib, "opencv_imgcodecs300d.lib")  
#pragma comment(lib, "opencv_imgproc300d.lib") //line, circle  
#pragma comment(lib, "opencv_core300.lib")  
#pragma comment(lib, "opencv_highgui300.lib")  
#pragma comment(lib, "opencv_imgcodecs300.lib")  
#pragma comment(lib, "opencv_imgproc300.lib") //line, circle  

//DEEP lib
#pragma comment(lib, "MareDeepDLL.lib")  

using namespace cv;
using namespace std;

void main()

 //DEEP Class
 MareDeepDll_B cDeep;

 //load model and structure
 cDeep.SetNet("lenet_test-memory-1.prototxt", "lenet_iter_10000.caffemodel");
 //gpu using on

 for (int i = 1; i <  14; ++i)
  // time check..
  unsigned long AAtime = 0, BBtime = 0;
  AAtime = getTickCount();

  //make file name
  char str[256];  
  sprintf_s(str, "%d.jpg", i);
  printf("%s\n", str);

  //img load and preprocessing
  Mat img = imread(str);
  resize(img, img, Size(28, 28));
  cvtColor(img, img, CV_BGR2GRAY);  

  vector< double> rV;
  //image and class num (caution!! class num is dependented by learning condition.) lenet is classify one number in 10 digits.
  rV = cDeep.eval(img, 10);

  //result out
  for (int i = 0; i <  rV.size(); i++) {
   printf("Probability to be Number %d is %.3f\n", i, rV[i]);   

  // processing time check.
  BBtime = getTickCount();
  printf("%.2lf sec / %.2lf fps\n", (BBtime - AAtime) / getTickFrequency(), 1 / ((BBtime - AAtime) / getTickFrequency()));

  namedWindow("test", 0);
  imshow("test", img);


Lenet model was used to test the deep learning classification.
Many other models are introduced on github model zoo.
You can apply other case, on code cDeep.SetNet("lenet_test-memory-1.prototxt", "lenet_iter_10000.caffemodel"); , first param means model structure and second param means the result of deep learning.

If you request to Google Plus to me, I will send the dll with the application code(project).


  1. Hii, can you post the training process? I have some problems to set training parameters

    1. I plan to post the code about learning method of deep learning.
      However, it is difficult for the parameter settings.
      because it require understanding of deep learning algorithm.
      I think tensor flow is good to study deep learning.
      I plan to post later about tensor flow.

      Can you open your learning data and parameter?
      Then I also try learning.

      Thank you.


Post a Comment

Popular posts from this blog

OpenCV Stitching example (Stitcher class, Panorama)

(OpenCV Study) Background subtractor MOG, MOG2, GMG example source code (BackgroundSubtractorMOG, BackgroundSubtractorMOG2, BackgroundSubtractorGMG)

Example source code of extract HOG feature from images, save descriptor values to xml file, using opencv (using HOGDescriptor )

Real-time N camera stitching Class.

8 point algorithm (Matlab source code) / The method to get the Fundamental Matrix and the Essential matrix

Optical Flow sample source code using OpenCV

Video Stabilization example source code, (using cvFindHomography, cvWarpPerspective functions in openCV)

(OpenCV Study) calcOpticalFlowFarneback example source code ( dense optical flow )

yuv422(YUYV) to RGB and RGB to yuv422(YUYV), (Using OpenCV and TBB)

(C, C++) TinyXML , xml read & write