Two-channel cv::Mat object filled with row-col indexes without for cycle

Question

I want to create a two-channel matrix in OpenCV, whose values are the corresponding pair of row and column indexes. I can easily do that in the following way:

for (int i = 0 ; i< img_height ; ++ i){
    for (int j = 0 ; j < img_width ; ++ j){
        src.ptr<Point2f>(i)[j] = Point2f(j,i);
    }
}

I wonder if there is a way in OpenCV to initialize such a matrix in a faster and more compact way, without necessarily using this element-wise approach. I searched on the documentation, but I didn't find anything that could help me for this purpose.

I ask that because I do need my application to be faster, so I'm looking for any possible improvements I can apply on my code.

Thanks in advance

Answer 1

There's no builtin function to to this. You can eventually mimic the Matlab function meshgrid using repeat, but in this case is going to be much slower.

You can however improve a few things:

• get the pointer to the beginning of the line out of the inner loop, since it will be the same for each line.
• avoid to create a temporary object to store values.
• I think you swapped i and j.

Have a look at this code:

Mat2f src(img_height, img_width);

for (int i = 0; i < img_height; ++i) {
    Vec2f* ptr = src.ptr<Vec2f>(i);
    for (int j = 0; j < img_width; ++j) {
        ptr[j][0] = i;
        ptr[j][1] = j;
    }
}

This snippet is a little faster (time in ms):

@MarcoFerro:    1.22755
@Miki:          0.818491

Testing code:

#include <opencv2/opencv.hpp>
using namespace cv;
using namespace std;

int main()
{
    int img_height = 480;
    int img_width = 640;

    {
        Mat2f src(img_height, img_width);

        double tic = double(getTickCount());
        for (int i = 0; i< img_height; ++i){
            for (int j = 0; j < img_width; ++j){
                src.ptr<Point2f>(i)[j] = Point2f(i, j);
            }
        }

        double toc = (double(getTickCount()) - tic) * 1000.0 / getTickFrequency();
        cout << "@MarcoFerro: \t" << toc << endl;
    }
    {
        Mat2f src(img_height, img_width);

        double tic = double(getTickCount());
        for (int i = 0; i < img_height; ++i) {
            Vec2f* ptr = src.ptr<Vec2f>(i);
            for (int j = 0; j < img_width; ++j) {
                ptr[j][0] = i;
                ptr[j][1] = j;
            }
        }

        double toc = (double(getTickCount()) - tic) * 1000.0 / getTickFrequency();
        cout << "@Miki: \t\t" << toc << endl;
    }

    getchar();
    return 0;
}