# mat.reshape not working

Hello I have a matrix M 3x3

I use reshape(0,9) so the M.at(0,1) should be the 4th element but it becomes the 2nd element. Help please. If i am wrong then please tell me how do i convert a 2D matrix into a 1D column vector.

EDIT: I dunno how to show the code. Its too big

My matrix is 92 x 112. some of the data below.

-40 -39

-44

....

In matlab if i reshape it becomes

-40

-44

...

-39

-39 becomes the 93rd element. In opencv I reshape it becomes

-40

-39

....

-39 becomes the SECOND element and i dunno where the -44 goes to. I used M.reshape(0,92*112). Matlab is giving me the right answer

Here is the short piece of code that I have

        cout << Im.at<float>(0,0) << endl;
cout << Im.at<float>(0,1) << endl;
waitKey(0);

// Reshape to column vector -- Wrong
Mat temp;
Im.reshape(0, 92*112).copyTo(temp);

cout << temp.at<float>(0,0) << endl;
cout << temp.at<float>(1,0) << endl;


What I am getting above is that the elements in (0,0) and (0,1) are equal to (0,0) and (1,0) after being reshaped. Im(0,1) should be equal to temp(113,0) or temp(93,0) but I am not getting such values. Is there another function to reshape a 2D Mat ?

EDIT: Okay it seems like my understanding of the reshape function was totally wrong. Thanks again berak. I had thought OpenCV would go 'down' the array but it seems like OpenCV would move from top left to bottom right

edit retag close merge delete

( 2016-03-24 02:52:46 -0500 )edit

I have tried it and it works for small matrices but not big ones. is there another function that I can try

( 2016-03-24 03:04:04 -0500 )edit
• "so the M.at(0,1) should be the 4th element" -- why do you think so ?
• "but it becomes the 2nd element" -- that's correct for the 3x3 case, for the 1x9 case, you're out of bounds. c++ starts indexing from 0, matlab from 1.
( 2016-03-24 03:04:41 -0500 )edit
1

be extra careful, when moving between opencv/c++ and matlab. again, there's a lot of assumptions:

• opencv/c++ starts indexing from 0, matlab/fortan from 1
• opencv Mat's are row-major (like in memory), matlab is col-major
• if there are channels, they are interleaved in opencv, but consecutive in matlab
• ... and lots more, that i can't remember
( 2016-03-24 03:56:43 -0500 )edit

Sort by ยป oldest newest most voted

your reshape is correct, but then, you seem to have trouble with different indexing conventions between c++ in general, and matlab.

please, when in doubt, you should run a debug build, a lot of important assertions are suppressed in release.

here's some code to illustrate it:

Mat_<float> m(3,3); m << 1,2,3,4,5,6,7,8,9;
cerr << m << endl;

[1, 2, 3;
4, 5, 6;
7, 8, 9]

cerr << m(0,1) << endl;
2

cerr << m(1,0) << endl;
4

Mat_<float> m2 = m.reshape(0,9);
cerr << m2 << endl;
[1;
2;
3;
4;
5;
6;
7;
8;
9]

// out of bounds:
cerr << m2(0,1) << endl;

OpenCV Error: Assertion failed ((unsigned)i1 < (unsigned)size.p[1]) in cv::Mat_<float>::operator (), file E:\code\opencv\build\install\include\opencv2/core/mat.
inl.hpp, line 1602

// this would be the correct way for a 1d col-mat:
cerr << m2(1,0) << endl;
2


## [edit:]

it seems, you want the transposed mat reshaped:

Mat mt = m.t();
cerr << mt << endl;
[1, 4, 7;
2, 5, 8;
3, 6, 9]

cerr << mt.reshape(0,9) << endl;
[1;
4;
7;
2;
5;
8;
3;
6;
9]

more

Official site

GitHub

Wiki

Documentation