Still cant figure out face recognition under 8.7

[Chad Bersche]

Hi Charlie, et. al

I have an Unconfirmed tag, but it has zero images in it, even after 
multiple "recognize faces" runs.

Once I did the initial scan all images, the Unknown tag had ~6000 images 
in it.  I classified probably ~800 of those naming individuals 
correctly.  I then re-ran the process as you indicated using "Recognize 
faces only" on all of my albums.  I watched it as it zoomed thru images, 
showing faces in the process of those in the image that it'd previously 
detected.  At the conclusion of the face detection, which took several 
minutes, it came back stating it scanned over 5,533 images and found 
ZERO faces.  Nothing moved from Unknown to Unconfirmed.  This is the 
same outcome every single time I've tried this.

So, what now?  I've been tempted to try to re-create the training data, 
but haven't done so yet.

   -- Chad


On 7/10/2025 10:07 PM, Charlie Gorichanaz wrote:
> Chad,
>
> Do you have a tag called "Unconfirmed" alongside your "Unknown" tag? I 
> believe that's where the detected-and-potentially-matched faces should 
> end up, waiting for you to confirm or reject them.
>
> The doc describes this under the Face recognition section, "To start 
> this process, go to the workflow tab, select *Recognize faces only*, 
> and then click *Scan Collection For Faces*. The process will tag 
> similar *detected* faces with the appropriate People tag from your tag 
> tree, and the corresponding thumbnail will disappear from the 
> *Unknown* tag selection. The thumbnail will appear in the 
> *Unconfirmed* tag section and will be added to the suggested face tag 
> section."
>
> —
> Charlie Gorichanaz
>
>
> On Thu, Jul 10, 2025 at 7:14 PM Chad A. Bersche <chad at bersche.com> wrote:
>
>     I've been baffled by this as well. I'm running on Windows 11, DigiKam
>     8.7, and downloaded all the modules.  I let it grind thru a bunch
>     of my
>     libraries overnight.  It resulted in about 6000 images in the Unknown
>     category.  I spent a a couple hours assigning names to people in that
>     list (sometimes up to 200-300 images).  I still have 5500 in the
>     Unknown
>     bucket after classifying several hundred.
>
>     After doing that, I went to run the "Recognize faces only" with
>     Accuracy=7, Search in set to "All Albums" and Tags unchecked. Then
>     when
>     I do the "Scan collection for faces" nothing changes, and all the
>     same
>     Unknown images stay in the same Unknown bucket.
>
>     What am I doing wrong?  I had hoped that tagging a good number of
>     faces
>     before doing the recognition would improve its accuracy, but it
>     seems to
>     just want me to do all the assignment work, which I kinda don't
>     want to do.
>
>     I'm using an external SQL database, with the database running on
>     Linux,
>     because I do use the same image libraries across Windows clients.
>
>     Thanks for any help/suggestions!
>
>        -- Chad
>
>     On 7/8/2025 10:43 PM, Gilles Caulier wrote:
>     > Hi,
>     >
>     > You need to first detect faces on albums before you recognize them.
>     > After detecting, manually tag a few faces with names, at least a few
>     > ones are enough to teach the AI for an automated recognition.
>     >
>     > Finally, run a recognition process and look as AI recognizes similar
>     > faces already registered in the model but not yet tagged in the
>     > database.
>     >
>     > The workflow is explained in the online doc :
>     > https://docs.digikam.org/en/left_sidebar/people_view.html
>     >
>     > Best
>     >
>     > Gilles Caulier
>     >
>     > Le mer. 9 juil. 2025 à 02:08, robert lindsay
>     > <robert.lindsay at gmail.com> a écrit :
>     >> This is under intel mac os and I installed the modules it asked
>     for in installation. Under face recognition their are boxes for
>     rebuilding and training data, but i can't find on to start face
>     recognition on albums
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.kde.org/pipermail/digikam-users/attachments/20250711/fc3b1334/attachment.htm>