operator* (Matx<_Tp,m,n>, Vec<_Tp, n>) should return Vec<_Tp,m> (Patch #1830)

Added by Yusuke Kameda almost 13 years ago. Updated almost 13 years ago.

Status:Done Start date:2012-04-21
Priority:Normal Due date:
Assignee:Vadim Pisarevsky % Done:

0%
Category:core
Target version:2.4.0
Affected version: Operating System:
Difficulty: HW Platform:
Pull request:

Description

Hello.
I found an improvement and a bug which is resolved by it.

In OpenCV 2.3.1, there is only a function

1Matx<_Tp, m, n> operator * (const Matx<_Tp, m, l>& a, const Matx<_Tp, l, n>& b);

and thus operator*(Matx, Vec) returns Matx<_Tp,m,1>.

Therefore, some codes like the following are illegal because Vec does not have a constructor from Matx<_Tp, m, 1>.

1Matx33d m33;
2Vec3d v3;
3Vec3d v = m33 * v3; // compile error at constructing Vec

Moreover, there is a bug in operator*(Matx, Scalar) because Scalar does not have a constructor from Matx<_Tp, 4, 1>.

1template<typename _Tp> static inline
2Scalar operator * (const Matx<_Tp, 4, 4>& a, const Scalar& b)
3{
4    return Scalar(a*Matx<_Tp, 4, 1>(b[0],b[1],b[2],b[3]));
5}


1Scalar s1;
2Matx44d m44;
3m44 * s1; // bug! compile error in operator*

If operator*(Matx, Vec) returns Vec, these problems are resolved.
Because Scalar has a constructor from CvScalar, which can be converted from Vec.

The following is a patch.

1template <typename _Tp, int m, int n> static inline
2const Vec<_Tp, m> operator*(const Matx<_Tp, m, n> &a, const Matx<_Tp, n, 1> &b) {
3  return Vec<_Tp, m>(Matx<_Tp, m, 1>(a, b, Matx_MatMulOp()), 1, Matx_ScaleOp());
4}

I have tested it in OpenCV 2.3.1, but not in 2.4 beta.
So if you accept this proposal, please test it.

Thank you for your kind cooperation.

Associated revisions

Revision 75b1788c
Added by Vadim Pisarevsky almost 13 years ago

added "Matx*Vec -> Vec" operator (ticket #1830)

Revision 87888984
Added by Vadim Pisarevsky almost 13 years ago

added Matx<4,4>*Scalar operator (ticket #1830)

Revision 63b3db53
Added by Roman Donchenko over 11 years ago

Merge pull request #1830 from SpecLad:config-includes

History

#1
Updated by Andrey Kamaev almost 13 years ago

  • Description changed from Hello. I found an improvement and a bug which is resolved by it. In OpenCV 2... to Hello. I found an improvement and a bug which is resolved by it. In OpenCV 2... More
  • Assignee set to Vadim Pisarevsky

#2
Updated by Vadim Pisarevsky almost 13 years ago

thanks! The operator has been added in r8175

  • Status changed from Open to Done

#3
Updated by Yusuke Kameda almost 13 years ago

Thank you very much!

However, I think the bug 

Scalar s1;
Matx44d m44;
m44 * s1; // bug! compile error in operator*

has not been fixed yet.
It is just that
Scalar(a*Matx<_Tp, 4, 1>(b[0],b[1],b[2],b[3]));

is illegal, because the code (a*Matx<_Tp, 4, 1>) still calls
Matx<_Tp, m, n> operator * (const Matx<_Tp, m, l>& a, const Matx<_Tp, l, n>& b)

instead of the added operator,
Vec<_Tp, m> operator * (const Matx<_Tp, m, n>& a, const Vec<_Tp, n>& b)

doesn't it?

From the added operator, I think the following is a patch for the bug.

template<typename _Tp> static inline
Scalar operator * (const Matx<_Tp, 4, 4>& a, const Scalar& b)
{
    Matx<double, 4, 1> c(a, b, Matx_MatMulOp());
    return reinterpret_cast<const Scalar&>(c);
}

So please check it.

Best regards.

#4
Updated by Vadim Pisarevsky almost 13 years ago

thanks! added in r8191

Also available in: Atom PDF