Bug 472645

Summary:	Face Recognition does not recognize any faces even after rebuilding and retraining all training data.
Product:	[Applications] digikam	Reporter:	Stewart Bodzin <Stewart.Bodzin>
Component:	Faces-Recognition	Assignee:	Digikam Developers <digikam-bugs-null>
Status:	RESOLVED DUPLICATE
Severity:	normal	CC:	metzpinguin, pschonmann
Priority:	NOR
Version:	8.2.0
Target Milestone:	---
Platform:	Microsoft Windows
OS:	Microsoft Windows
Latest Commit:		Version Fixed In:
Attachments:	attachment-3958260-0.html

Description Stewart Bodzin 2023-07-26 05:26:45 UTC

Face recognition is really broken.  Upgraded to 8.2.0 and no faces seem to get correctly recognized.  I cleaned up any mis-identified faces and also did a full Finger-prints rebuild and did a Detect and recognize faces from the maintenance menu.  Selected clear and rebuild all training data.  Face detection continues to run ok but the face recognition in all modes with and without YOLO is totally inaccurate on all faces.  I have many faces identified for each unique person and did the rebuild training.  However, even on a single image, selecting the context menu to recognize faces finds no faces on all photos I have tested.  Common faces are also not found on any single photo.  Something is seriously wrong.  Thank you.

Comment 1 Maik Qualmann 2023-07-26 05:51:00 UTC

Have you narrowed down the face detection/recognition  to albums or tags in the People sidebar in the settings. Unfortunately, users wanted us to turn face recognition back to select albums as well. Select all albums and no tags.

Maik

Comment 2 Maik Qualmann 2023-07-26 07:47:48 UTC

The problem is, the more confirmed faces you have from one person, the more watered down the result and you end up not recognizing faces. We have to think about limiting it.

Maik

*** This bug has been marked as a duplicate of bug 431797 ***

Comment 3 Stewart Bodzin 2023-07-27 05:45:25 UTC

Created attachment 160558 [details]
attachment-3958260-0.html

Hey Maik. I agree with your musings about this problem. The process should have two types of confirm. 1. Confirm the face is correctly identified or 2. Confirm the face is correctly identified and commit that face image to the training data set for the neural net. This way we'd only need to identify a handful of good face identifiable pictures during the beginning of a new face and then do confirms just to tag the picture but not to pollute the training dataset.Just my 2 cents.Sent from my Galaxy
-------- Original message --------From: Maik Qualmann <bugzilla_noreply@kde.org> Date: 7/26/23  1:39 AM  (GMT-08:00) To: Stewart.Bodzin@usa.net Subject: [digikam] [Bug 472645] Face Recognition does not recognize any faces even
  after rebuilding and retraining  all training data. https://bugs.kde.org/show_bug.cgi?id=472645Maik Qualmann <metzpinguin@gmail.com> changed:           What    |Removed                     |Added----------------------------------------------------------------------------         Resolution|---                         |DUPLICATE             Status|REPORTED                    |RESOLVED--- Comment #2 from Maik Qualmann <metzpinguin@gmail.com> ---The problem is, the more confirmed faces you have from one person, the morewatered down the result and you end up not recognizing faces. We have to thinkabout limiting it.Maik*** This bug has been marked as a duplicate of bug 431797 ***-- You are receiving this mail because:You reported the bug.