Starting with Face Detection: Wait for new tag? And how to change employed model?

[Tonio Kroeger]

Wow! Thanks a lot!! 🥰

On Fri, Jan 17, 2025 at 11:28 AM Michael Miller <michael_miller at msn.com>
wrote:

> Hi Tonio,
> I hope you’re having a good day!
>
> I  came across "warnings" here in the mailing list that face recognition
> might degenerate when tagging too many blurred or otherwise unfavorable
> pictures. At the same time a new tag was proposed, in my words "confirmed
> person but do not use it for improving the face recognition". So does it
> make sense to wait for this new tag? That is to say for the time being I
> only tag the good pictures and leave the blurry ones "unconfirmed” .
>
> We added code to automatically filter out blurry faces so they are not
> used in face training.  You can still tag the faces and see them in digiKam
> as normal, but blurry faces are not used to train the recognition model
> (it’s called the classifier).  The code is still very experimental, but
> should help improve the results when digiKam is recognizing new faces.
>
> My first question: Is it a good idea to ail or should I just tag
> everything?
>
> The next AppImage build will include the blurry face filter code.  I would
> tag only good faces now, and then you can tag everything after you download
> and install the next build.  The next build should be available in a day or
> two.
>
> Second question: The face recognition settings under People -> Settings
> only show "Accuracy", nothing to choose a model... Some people here wrote
> about an improved models like "SFace" (Mail from Steven Robbins, January
> 12th, 2025). Where do I find the alternative models?
>
> SFace is the only model digiKam uses now. We removed the older models
> since they were not as reliable.
>
> Cheers,
> Mike
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.kde.org/pipermail/digikam-users/attachments/20250117/357ee445/attachment.htm>