mark92m's profile - activity

I am trying to fill a struct with Mat objects to pass to a thread.

   struct thread_data{
           int thread_id;
           int socket;
           uchar* receiveBuffer;

           // MAT IMAGE HEADERS:
           Mat _leftImage;
           Mat _rightImage;
           Mat _leftDepth;
           Mat _rightDepth;
           vector<Mat> _results;
      };

Then I create an array of structs like this:

  struct thread_data td[MAX_THREADS];

Following, each struct gets populated, where the Mats in the struct are filled with Mats saved in Vectors :

for(i = 0; i < numThreads ; i++){
  td[i].thread_id = i;        
  td[i].socket = connected;         
  td[i].receiveBuffer = threadRecBuffer[i];

  td[i]._leftImage = leftImages[i];   
  td[i]._rightImage = rightImages[i]; 
  td[i]._leftDepth = leftDepth[i];    
  td[i]._rightDepth = rightDepth[i];     
  td[i]._results = resultsVector;           
   }

Works well up to 9 iterations, and starts giving the error at the 10th iteration.

I get a segmentation fault 11 as soon as the first Mat is trying to get assigned.

Has anyone got any experience with passing Mats in structs ?

Note: All images exist and are being printed before hand.

Thanks in advance.

In my code , I use sockets to offload processing of a Mat Image onto another machine, and wait for the result to be read back. So in end up with resulting the Mat Image.

I use fork() to create a child process to take care of the sending and receiving of this Mat image.

Mat image = imread( "image_source" );
Mat result;

fork()  
if( child_process ){

      // OVER SIMPLIFYING:
      open socket();
      send( image.data );
      receive( result.data );
}      

if( parent_process ){ 
    wait();
    imshow(  result ) ;
}

The socket programming works fine and the child thread gets the correct information (verified). Obviously I need to access the resulting image from the parent thread and NOT the child thread only.

The parent thread keeps wait indefinitely, as Mat result is never populated.

Do I need to use some form of shared memory ? Or is that bad practice when using threads ?

I have an OpenCV application which I would like to speed up using several GPUs (located in a linux machine cluster).

My approach was to create an OpenCL kernel to execute on a single GPU, and then extend this performance by using VirtualCL to use the network-attached devices.

My question is, is it possible to run an OpenCL kernel that somehow calls the OpenCV Library ?

clBuildProgram() allows you to specify compiler options which include the include file directories. The header file includes, should look something like: -I myincludedir1 -I myincludedir2

similar to:

g++ -I/usr/local/include/opencv -I/usr/local/include/opencv2 -L/usr/local/lib/ -g -o binary  main.cpp -lopencv_core -lopencv_imgproc -lopencv_highgui -lopencv_ml -lopencv_video -lopencv_features2d -lopencv_calib3d -lopencv_objdetect -lopencv_contrib -lopencv_legacy -lopencv_stitching

Has anyone ever managed to create an OpenCL kernel that uses OpenCV ? Or is there any simpler way ? (without using UMat )

I want to process a set of images and display them in real time in order to time the frames per second with which the code executes.

    for(int i = 0 ; i  < numOfImages ; i ++ ){

        * DO SOMETHING WITH IMAGE * 

        imshow("Frame", image); 
        waitKey(1);
     }

Is the waitKey(1) the most efficient way to display the succession of frames as fast as possible ? Or is there a better way ?

Thanks!

Indeed it was quicker, thank you both for your help.

I need to use a warping algorithm. Given an image (the Mat mentioned above), each pixel has a depth value ( stored in another Mat). The algorithm takes a set of coordinates and warps them to new coordinates.

So for each row of this matrix, I shall calculate a new set of values.

I wish to parallelise this process. Hence why i have isolated it into a separate process.

Hi,

In my code I need to create a matrix filled with all the possible pixel coordinates of an image ( Mat ).

ex.for a 1024x768 image: [ 0, 0] [ 0, 1] [ 0, 2] . . . [1024, 768]

Is there a more efficient way of doing this than:

void indexMatrix(Mat &matrix, int height, int width){

int cols = height*width;
int index = 0;

matrix = Mat::ones(cols, 2, CV_8UC1);

for( int col = 0; col < width; ++col ) {
    for (int row = 0; row < height; ++row) {
        matrix.at<uchar>(index,0) = col;
        matrix.at<uchar>(index,1) = row;

        index++;
    }
}

... cause I've been looking "deeply" at it for quite some time today, but I'm still getting used to how OpenCV operates. I think i understand what you mean.

So you suggest first dividing all elements by 255, then negating it, then adding 1, then using the mul() function , and so on ... for each step of the algorithm ?

Does anyone know of a more efficient OpenCV practice to perform the following operation on each Mat element ? Each element undergoes the same function.

Mat calcDepthValues(Mat &depth){

Mat realDepth= depth.clone();

Mat_<double>::iterator it = realDepth.begin<double>();
Mat_<double>::iterator itend = realDepth.end<double>();

for (; it != itend; ++it) {

    (*it) = 448.251214 +  ((1- (*it)/255)  * 10758.02914);

}
return realDepthValues;

I have edited my question to include a more complex example.

Thanks for the very beneficial tip.

But not all my operations are this simple.

I Have cases when i require a pixel(x, y) from Mat A, another pixel(x, y) from Mat B, and input these pixel values into a specific algorithm. Then I store the result into pixel(x, y) of Mat C.

I don't know how to proceed with such complex situations.

Would it be more helpful if I add some example code ?

Thanks

Hi, I am using OpenCV 3.0 .

In my serial implementation I use the Mat_::iterator quite often to access and edit each pixel of an image (As in the example below).

What is the best practice to accelerate these calculations ? Unfortunately I cannot find ways to access particular pixels in a UMat matrix.

Mat createFGMask(Mat &depthMap){
Mat fgMask = depthMap.clone();

// obtain iterator at initial position
Mat_<uchar>::iterator it= fgMask.begin<uchar>();
// obtain end position
Mat_<uchar>::iterator itend= fgMask.end<uchar>();

// loop over all pixels
for ( ; it!= itend; ++it) {
    //IF VALUE GREATER THEN 120, SET TO FOREGROUND ( ie. 1):
    if( (*it)> 120 ) (*it)=255;
    else (*it)= 0;
}

return fgMask;

}

A more complex scenario where values of each pixel is required:

Mat pixelShifting(Mat &refImage, Mat &depthValue, ){

Mat warpedImage = Mat::zeros( 768, 1024, CV_8UC3 );

int height = refImage.rows;
int width = refImage.cols;

double newX, newY;
Mat newCoord;


for( int col = 0; col < refImage.cols; ++col ){
    for( int row = 0; row < refImage.rows; ++row ){

        //FIRST SHIFT BACKGROUND PIXELS:
        if( depthValue.at<uchar>(row,col) <= 120 ){

              // ALGORITHM THAT COMPUTES NEW PIXEL COORDINATES.
            newCoord = calcNewCoord(col, row, depthValues.at<double>(row,col));                                                                                  

            newX = newCoord.at<double>(0, 0);
            newY = newCoord.at<double>(1, 0);

            if( 0<= newY < height && 0<= newX < width ) {

                warpedImage.at<Vec3b>(newY, newX) = refImage.at<Vec3b>(row, col);

            }
        }
    }
}
return warpedImage;

}

This code basically takes 2 images, and populates a 3rd image with pixels of image1 by calculating new coordinates using values from image 2.