[digiKam-users] People Vs person tag trees

HaJo Schatz hajo at hajo.net
Thu Apr 11 04:14:50 BST 2019


On Tue, Apr 2, 2019 at 3:51 AM Timo Kaufmann <timokau at zoho.com> wrote:

> On 19-04-01 12:12, woenx wrote:
> > 1) you scan the pictures for faces (either from the "People" panel or by
> > right-clicking an album and scanning it for faces. You can either just do
> > the face detection, or also face recognition.
> >
> > 2) When the scan finishes. Go to the "Unknown" tag in the "People" panel.
> > New detected faces will be there, and you can name them. After naming
> them,
> > you'll find those faces under the name you used in the People panel.
>
> I have always wondered: Is there some way I can tag a face as "Unknown
> person"? As in "yes, that is a face but I don't know who that is".
> Removing the face-tag is not ideal (for example it will reappear on a
> full rescan) and naively tagging it as "Stranger" will throw off the
> recognition algorithm.
>
>
I agree to this issue, I assume that without such function the concept of
face recognition is rather useless for anything more than a few images.

One more question that just came up in this context: In my people tree
(digikam 5.9), I see both an 'Unknown' and 'Unconfirmed' tag. Not sure
whether 'Unconfirmed' is a legacy entry or whether it serves any purpose?

Logically, with my limited English skill, 'Unconfirmed' would be the faces
that are not known to the algorithm and should be confirmed by the user.
'Unknown' are people that are confirmed not known, I.e. strangers, and be
excluded from the AI learning in future runs. But as of 5.9, digikam places
newly detected, unidentified faces into 'Unknown".

Forgive me please if this has changed significantly in 6.0, I haven't had
the chance yet to test it since I like to stay with distro versions where
possible.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.kde.org/pipermail/digikam-users/attachments/20190411/f35b920f/attachment.html>


More information about the Digikam-users mailing list