Skip to content

Small inefficiency in seamless clone init #16168

Closed
Closed
Small inefficiency in seamless clone init#16168
Labels
affected: 3.4category: photogood first issue
Milestone
3.4.10
@dan-masek

Description

@dan-masek
dan-masek
opened on Dec 15, 2019

I have just been reading through the implementation of seamless cloning, and ran across those two bits of code:

Cloning::computeGradientX

opencv/modules/photo/src/seamless_cloning_impl.cpp

Lines 58 to 66 in 4c71dbf

else if (img.channels() == 1)
{
Mat tmp[3];
for(int chan = 0 ; chan < 3 ; ++chan)
{
filter2D(img, tmp[chan], CV_32F, kernel);
}
merge(tmp, 3, gx);
}

Cloning::computeGradientY

opencv/modules/photo/src/seamless_cloning_impl.cpp

Lines 79 to 87 in 4c71dbf

else if (img.channels() == 1)
{
Mat tmp[3];
for(int chan = 0 ; chan < 3 ; ++chan)
{
filter2D(img, tmp[chan], CV_32F, kernel);
}
merge(tmp, 3, gy);
}

It appears to me that the loop calculates the same thing 3x in a row. Wouldn't it make more sense to just run filter2D once, and either merge 3 shallow copies of the result, or outright use mixChannels?

Metadata

Metadata

Assignees

No one assigned

    Labels

    affected: 3.4category: photogood first issue

    Type

    No type

    Projects

    No projects

    Milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions