Face Recognition scope issues.

[George Koulomzin]

Hi,

I'm a relatively new user of DigiKam, and so far quite pleased with the 
program!

However I am struggling to master DigiKam's face detection and 
recognition.  Mostly, these issues relate to the set of images to which 
Recognition is to be applied.  Here are a couple of cases that have me 
puzzled:

 1.   Selected unknown faces do not get recognized.
     1. I have a (large) number of faces in the Unknown people tag.
     2. I select several face thumbnails of the same person.  This
        person's face has already been tagged 147 times (i.e. the tag
        tree shows 147 entries for this person).
     3. I press "recognize faces" in the workflow tab of the people view
     4. i press "scan for faces" in the Item menu. (I would have
        expected to see a recognize face menu entry, but there is none).
     5. I dialog appears saying the right number of images were scanned,
        but no faces were recognized, and they remain in the unknown tag.
     6. Why is this happening?
     7. the same thing happens if I select "recognize faces" from the
        context menu in step 4.
 2. I foolishly (in hindsight) moved a bunch of faces from the unknown
    tag to a temporary "test" tag that I created.
     1. I moved most of the tags in unknown to a new temporary tag
     2. I then did face recognition.
     3. I then moved the faces in the temp tag back to unknown
     4. when I tried to recognize those faces, they all suggested "temp"
        (makes sense in hindsight)
     5. I deleted the temp tag.
     6. when I tried to recognize those faces again, the still suggested
        "temp".
     7. Hence I presume that DK did not remove the face recognition data
        associated with the deleted tag....
     8. How can I make DK *forget* a face?  Just some.  Not all.
 3. A lot of the faces I have detected and recognized to date are of
    poor quality. For example, blurry, pixelated, taken from almost
    behind the person, etc.  While I am impressed with DK's ability to
    detect that these are indeed faces, I have read that poor quality
    training data make for bad recognition.
     1. for the purpose of finding people in my collection, I would like
        to tag even the bad images.
     2. for the purpose of good training data, it seems I should only
        tag good images
     3. How do I accomplish both?
     4. BTW, I believe that the "accuracy" setting (sensitivity <->
        specificity) is not well named.  With respect to detection, I
        would suggest "aggressiveness" (cautious <-> aggressive), or
        (false positives (fewer <-> more).
     5. I am not sure what this setting means w.r.t. recognition.
 4. In the "search-in" tab, if I specify albums and also specify tags,
    is the domain of the search the intersection of those two specifiers
    (and), or is it the union (or)?
 5. My goal is to detect and recognize faces incrementally, as I am
    processing a collection of images (approx 90k images) captured over
    many years.  I am hoping to effectively partition the unknowns by
    other selection criteria.

Running DK 8.2.0 on Win 11 Pro 23H2

Thanks in advance!

-- 
George Koulomzin
7 Bridge Hill Lane,
PO Box 781
Bridgehampton, NY, 11932
(m) 914-393-6179
(h) 631-537-4956
George at Koulomzin.com
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.kde.org/pipermail/digikam-users/attachments/20231209/7ab27119/attachment.htm>