Mat scaling affects only first channel (Bug #3227)

Added by Baris Demiroz over 11 years ago. Updated about 11 years ago.

Status:Cancelled Start date:2013-08-21
Priority:Normal Due date:
Assignee:Kirill Kornyakov % Done:

0%
Category:core
Target version:2.4.8
Affected version:2.4.0 - 2.4.6 Operating System:Any
Difficulty:Easy HW Platform:Any
Pull request:

Description

When a matrix constructed with Mat::ones() is scaled with * operator, it affects only the first channel.

The code to reproduce is below, the displayed image is all blue.

1Mat3b img = Mat3b::ones(600,600)*255;
2imshow("img", img);
3waitKey();

Associated revisions

Revision 9bba2443
Added by Vadim Pisarevsky over 10 years ago

Merge pull request #3227 from PhilLab:patch-3

History

#1
Updated by Baris Demiroz over 11 years ago

If this is the intended behavior it should be in the documentation. I couldn't find it in the docs, of course there is a chance I overlooked it.

#2
Updated by be rak over 11 years ago

problem already starts here:

Mat3b img = Mat3b::ones(3,3);
cerr << img << endl;

[ 1, 0, 0, 1, 0, 0, 1, 0, 0;
1, 0, 0, 1, 0, 0, 1, 0, 0;
1, 0, 0, 1, 0, 0, 1, 0, 0]

or, for a change:

Mat img = Mat::eye(3,3,CV_8UC3);
cerr << img << endl;

[ 1, 0, 0, 0, 0, 0, 0, 0, 0;
0, 0, 0, 1, 0, 0, 0, 0, 0;
0, 0, 0, 0, 0, 0, 1, 0, 0]

#---------------------------------------------------------------------------------------

tracked it down to core\src\matop.cpp, L1555 :

void MatOp_Initializer::assign(const MatExpr& e, Mat& m, int _type) const {
if( _type == -1 )
_type = e.a.type();

if( e.a.dims <= 2 )
        m.create(e.a.size(), _type);
    else
        m.create(e.a.dims, e.a.size, _type);
if( e.flags == 'I' && e.a.dims <= 2 )
        setIdentity(m, Scalar(e.alpha));
    else if( e.flags == '0' )
        m = Scalar();
    else if( e.flags == '1' )
        m = Scalar(e.alpha);
    else
        CV_Error(CV_StsError, "Invalid matrix initializer type");
}

as you can see, in both 'I' and '1' cases, Scalar(e.alpha) is used(which only initializes the 1st element).

it's a bit unclear to me, what the desired behaviour is here. should Mat::ones() and Mat::eye() only be used with 1 channel Mat's ? then that needs clarification.

using Scalar::all(e.alpha) instead would fix it otherwise:

#---------------------------------------------------------------------------------------

Mat3b img = Mat3b::ones(3,3);
cerr << img << endl;
[ 1, 1, 1, 1, 1, 1, 1, 1, 1;
1, 1, 1, 1, 1, 1, 1, 1, 1;
1, 1, 1, 1, 1, 1, 1, 1, 1]

Mat img = Mat::eye(3,3,CV_8UC3);
cerr << img << endl;
[1, 1, 1, 0, 0, 0, 0, 0, 0;
0, 0, 0, 1, 1, 1, 0, 0, 0;
0, 0, 0, 0, 0, 0, 1, 1, 1]

#3
Updated by Dinar Ahmatnurov over 11 years ago

  • Status changed from New to Open

#4
Updated by Dinar Ahmatnurov over 11 years ago

"using Scalar::all(e.alpha) instead would fix it otherwise" as for me it sounds good, could you please open PR on github with your changes? Kirill, any additions?

  • Assignee changed from Vadim Pisarevsky to Kirill Kornyakov

#6
Updated by Dinar Ahmatnurov over 11 years ago

perfect, 10x!

#7
Updated by Kirill Kornyakov over 11 years ago

  • Target version set to 2.4.7
  • HW Platform changed from x64 to Any
  • Operating System changed from Windows to Any
  • Pull request set to https://github.com/Itseez/opencv/pull/1331
  • Difficulty set to Easy

#8
Updated by Kirill Kornyakov over 11 years ago

@Berak, was it finally merged to the repo? I can see the branch b_3227_24 in your repo, but I'm not sure if it was merged...

#9
Updated by Alexander Smorkalov over 11 years ago

@Berak, was the problem solved? Was the solution merged to OpenCV repository?

  • Pull request deleted (https://github.com/Itseez/opencv/pull/1331)
  • Affected version changed from 2.4.6 (latest release) to 2.4.0 - 2.4.6

#10
Updated by Alexander Smorkalov over 11 years ago

@Berak, was the problem solved? Was the solution merged to OpenCV repository?

#11
Updated by Alexander Smorkalov over 11 years ago

  • Target version changed from 2.4.7 to 2.4.8

#12
Updated by be rak over 11 years ago

sorry, it took so long, finally found out, how to do a proper 2.4 pr

i assumed, setIdentity() should behave the same way now, right ?

https://github.com/Itseez/opencv/pull/1900

#13
Updated by Ilya Lavrenov about 11 years ago

as you can see from https://github.com/Itseez/opencv/pull/1900 this behaviour of Mat::ones() is expected, closed.

  • Status changed from Open to Cancelled

Also available in: Atom PDF