Bug 262585 - A single face is detected twice
Summary: A single face is detected twice
Status: RESOLVED FIXED
Alias: None
Product: digikam
Classification: Applications
Component: Faces-Detection (show other bugs)
Version: 2.0.0
Platform: Compiled Sources Linux
: NOR normal
Target Milestone: ---
Assignee: Digikam Developers
URL:
Keywords:
Depends on:
Blocks:
 
Reported: 2011-01-08 21:10 UTC by Milan Knížek
Modified: 2017-07-26 18:00 UTC (History)
1 user (show)

See Also:
Latest Commit:
Version Fixed In: 2.5.0
Sentry Crash Report:


Attachments
Multiple detection of a single face (9.36 KB, image/jpeg)
2011-01-08 21:11 UTC, Milan Knížek
Details

Note You need to log in before you can comment on or make changes to this bug.
Description Milan Knížek 2011-01-08 21:10:22 UTC
Version:           2.0.0 (using KDE 4.5.4) 
OS:                Linux

After scan of collection, a face is shown twice in /People/Unknown. See the attachment - the "face boxes" overlap.

Reproducible: Always



Expected Results:  
Each face should be detected only once - that is the "face boxes" should never overlap.

See the attached screenshot.
Comment 1 Milan Knížek 2011-01-08 21:11:32 UTC
Created attachment 55746 [details]
Multiple detection of a single face
Comment 2 Marcel Wiesweg 2011-01-11 13:52:02 UTC
We do have overlap detection, only though when one face is already set and a new one is added. For faces added in one run, there is overlap detection in libface IIRC.
These do overlap, but not so much. The problem: Which face is the right one? Impossible to know! If the lower frame, probably fooled by the moustache, is added first, we'll have made a mistake.
Comment 3 Milan Knížek 2011-01-22 18:42:01 UTC
I get your point, Marcel. Then it is a question, what is easier from the user's viewpoint: (a) digiKam keeps both detected faces (though overlapped) and the user manually removes one of them as a false positive or (b) digiKam chooses the best fit and in case it is the wrong one, the user moves the frame to better fit the face.

I would probably opt for the latter solution, since there are chances that digiKam did the right thing and user does not have to interact at all.
Comment 4 Marcel Wiesweg 2011-01-24 16:03:29 UTC
I need to check if we can get "scoring" information from libface. For the normal detection, there is no such information; but maybe from the verification state.
Comment 5 caulier.gilles 2011-12-14 13:53:57 UTC
Milan, 

it still valid using digiKam 2.4 ?

Gilles Caulier
Comment 6 Milan Knížek 2011-12-16 19:09:34 UTC
Thanks for pushing to re-test, the bug does not appear anymore with today's GIT (on the same test album). (It still existed in 2.3.)