Member variable u of cv::Mat is still NULL

C++
1

OpenCV version 4.5.4

When a cv::Mat variable is created in a certain constructor, u, a member of cv::Mat, becomes NULL.
The following code is a simple example.

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

int main() {
    cv::Mat a = cv::Mat::zeros(3, 3, CV_8UC3);
    cv::Mat b = cv::Mat(a.rows, a.cols, CV_8UC3, a.data);

    std::cout << a << std::endl;
    std::cout << b << std::endl;
    std::cout << a.u << std::endl;
    std::cout << b.u << std::endl;
    return 0;
}

The results of this program are as follows.

[  0,   0,   0,   0,   0,   0,   0,   0,   0;
   0,   0,   0,   0,   0,   0,   0,   0,   0;
   0,   0,   0,   0,   0,   0,   0,   0,   0]
[  0,   0,   0,   0,   0,   0,   0,   0,   0;
   0,   0,   0,   0,   0,   0,   0,   0,   0;
   0,   0,   0,   0,   0,   0,   0,   0,   0]
00000203AC74C3A0
0000000000000000

Is this a constructor bug?
Also, is there any problem with u being NULL?

Thanks.

2

what is your interest in this member variable?
you have expectations of this member variable. what do you base these expectations on?

these are important questions, not rhetorical.

https://docs.opencv.org/4.x/d3/d63/classcv_1_1Mat.html#a2742469fe595e1b9036f60d752d08461

you must reproduce any issues with the current release or git master. we will not debug anything that could have been fixed in the meantime. 4.5.4 is ancient.

3

I’m wondering if the reference counter works correctly when u is NULL.

I found this problem while debugging the Windows bitmap to cv::Mat conversion process.
Then I realized that u contains the cv::Mat reference counter.

In my current environment, it is difficult to try the latest version of OpenCV right away.
I will verify as soon as possible.

4

I have verified this with version 4.10.0 and it outputs similar results.

I also found that b->u is not NULL in the following code.

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

int main() {
    cv::Mat a = cv::Mat::zeros(3, 3, CV_8UC3);
    cv::Mat b = cv::Mat(a.rows, a.cols, CV_8UC3);

    b.data = a.data;

    std::cout << "a = " << a << std::endl;
    std::cout << "b = " << b << std::endl;
    std::cout << "a->u = " << a.u << std::endl;
    std::cout << "b->u = " << b.u << std::endl;
    return 0;
}

a = [  0,   0,   0,   0,   0,   0,   0,   0,   0;
   0,   0,   0,   0,   0,   0,   0,   0,   0;
   0,   0,   0,   0,   0,   0,   0,   0,   0]
b = [  0,   0,   0,   0,   0,   0,   0,   0,   0;
   0,   0,   0,   0,   0,   0,   0,   0,   0;
   0,   0,   0,   0,   0,   0,   0,   0,   0]
a->u = 000001F31A6F65E0
b->u = 000001F31A6F5FB0

When creating a cv::Mat that shares a pointer to data, what is the recommended writing style?

5

recommended? to leave u alone. that’s internal plumbing.

I linked the docs for you earlier. Now I’ll quote them:

UMatData* cv::Mat::u

interaction with UMat

that is not something you should touch, not ever.

I asked why you thought you should touch that. I haven’t received an answer yet.

6

I don’t think u should be manipulated by the programmer; however, I am unsure if u being NULL is correct in this context.

As I mentioned above, I am creating a conversion process between a Windows bitmap and cv::Mat, which requires a pointer to the data to be shared between them.

However, when I define cv::Mat using the constructor cv::Mat(rows, cols, type, *data), the u field, which should contain the matrix metadata, is NULL.

What problems might this cause?
For example, the reference counter within u might have an incorrect value, leading to issues with memory not being freed correctly.

7

if you want to manipulate the same memory, then yes, you’d use a Mat that doesn’t own its memory. it cannot “manage” that memory.

you can always just copy that Mat. then you have a new Mat that does own its memory. that copy should behave in all the expected ways.

reference counting is a Mat-internal feature that only works on “owned” backing memory, i.e. backing memory that was created by a Mat. this is special. it’s not “arbitrary” memory you pass in as a pointer and size. it’s an actual instance of a class provided by OpenCV.

8

Thank you. Understood.

So when sharing data, the memory management is not done by cv::Mat, but by the object to which the data is shared.