Bug 495608

Summary:	Refine faces capability
Product:	[Applications] digikam	Reporter:	Tom Cunningham <twcunningham627>
Component:	Faces-Workflow	Assignee:	Digikam Developers <digikam-bugs-null>
Status:	RESOLVED FIXED
Severity:	wishlist	CC:	michael_miller
Priority:	NOR
Version First Reported In:	8.5.0
Target Milestone:	---
Platform:	Microsoft Windows
OS:	Microsoft Windows
Latest Commit:		Version Fixed In:	8.6.0
Sentry Crash Report:

Description Tom Cunningham 2024-10-30 20:10:05 UTC

SUMMARY
A feature to revisit faces and, for any faces that do not meet a percentage of recognizability, remove the face rectangle but retain the corresponding People tag.

STEPS TO REPRODUCE
1. Scan all faces to find the best baseline for each face (algorithmically determined).
2. Scan faces again and compare to baseline. Given a predetermined setting for percentage of recognizability using the baseline, if any face falls below the percentage, remove the rectangle identifying the face, but retain the corresponding People tag for that face (the person is still in the photo, but with no recognizable face, e.g. back turned).
3. Presumably then, the percentage of recognizability for face refinement will be the same as the setting for automatic face recognition.

OBSERVED RESULT


EXPECTED RESULT
Should only need to be run when a) the percentage of recognizability changes for automatic face recognition or b) when face tags have been used instead of People tags.

SOFTWARE/OS VERSIONS
Windows: 
macOS: 
(available in the Info Center app, or by running `kinfo` in a terminal window)
Linux/KDE Plasma: 
KDE Plasma Version: 
KDE Frameworks Version: 
Qt Version: 

ADDITIONAL INFORMATION

Comment 1 Michael Miller 2025-02-02 13:09:33 UTC

Hi Tom,
We've introduced a face image quality assessment (FIQA) check in 8.6.0.  FIQA checks for blurry, pixelated, and noisy faces, and removes them from being used for training, but keeps the tag and rectangle for the end user.

Cheers,
Mike

Comment 2 Tom Cunningham 2025-02-02 13:42:27 UTC

(In reply to Michael Miller from comment #1)
> Hi Tom,
> We've introduced a face image quality assessment (FIQA) check in 8.6.0. 
> FIQA checks for blurry, pixelated, and noisy faces, and removes them from
> being used for training, but keeps the tag and rectangle for the end user.
> 
> Cheers,
> Mike

This sounds great, Mike, thanks!