[digiKam-users] face tagging persistently ignore a face

[Marc Palaus]

    Adding blurry face regions to the "ignored" tag will cause other
    things to match those later.  I've seen real people that appear in
    multiple photos get "matched/recognized" against a name of "ignored"
    because they were ignored in previous runs.

Are you sure about that? I have never been suggested any face to the 
"Ignored" group (and I have tagged thousands of pictures). In the past, 
when you clicked on ignore, that "Ignored" would be saved in the 
metadata, but it was fixed last September 
(https://bugs.kde.org/show_bug.cgi?id=459537). I'd recommend you delete 
that Ignored face region from the pictures they currently have it, so 
those faces don't appear in other programs like Pigallery2. In theory, 
the ignored tag shouldn't be used for the face training (can a developer 
confirm that?).

As I said, I never had this issue. Clicking the red cross will just 
remove that face from digikam (although if the picture is scanned again, 
it will be re-detected), and the Ignore just hides them into their own 
category, but without writing anything to the pictures either.

El 2/3/23 a les 11:58, Travis Kelley ha escrit:
> Let me caveat this by saying I'm not an expert/developer, so I'm only 
> commenting on what I've seen.
>
> When you use the red X button it throws away that detected region 
> completely.  It doesn't save anything to the metadata of the photo.  I 
> don't believe anything gets saved in the database either (more on that 
> below).  It's as if that "detection" never happened.
>
> My understanding of the intention of detection is that you should only 
> need to run it on a photo once.  Hence there is a setting under 
> detection to only scan NEW photos.  You are not prevented from 
> rescanning a photo, but I believe the normal workflow would not be to 
> do that.  Hence using the red X button effectively makes it as if the 
> detection never happened and unless you specifically rescan a 
> previously scanned photo, that same region would not get detected again.
>
> Adding blurry face regions to the "ignored" tag will cause other 
> things to match those later.  I've seen real people that appear in 
> multiple photos get "matched/recognized" against a name of "ignored" 
> because they were ignored in previous runs. You need to be somewhat 
> careful in what you ignore as if you ignore someone important there is 
> a chance the next photo of that person will be "recognized" as the 
> "ignored" person.  You can always correct that match and tell digikam 
> that person is a real name if that happens.  From everything I can 
> tell, "ignored" is not treated as a special tag at all.  It 
> effectively creates a name of "ignored" that gets treated just like 
> any other person name.  It gets written to photo metadata and I can 
> for instance look at all the people named ignored in PiGallery2 (which 
> reads the face metadata that digikam created).  I suppose it would be 
> fairly easy to strip all of those out at any time by deleting the 
> "ignored" name in digikam or using something like exiftool, but they 
> haven't caused me to much trouble other than philosophically the 
> creation of useless metadata.
>
> Someone with more knowledge feel free to correct me.  I'm only going 
> based on experience having used digikam for a while.
>
> On Thu, Mar 2, 2023 at 7:08 AM Thomas <sdktda at gmail.com> wrote:
>
>
>     On 2023-03-02 13:00, Travis Kelley wrote:
>     > Yes.  When you "ignore" a face, you are essentially marking that
>     face
>     > with a name of "ignore".  Other software will see the tag in the
>     > metadata of that photo.
>
>
>     Does this mean that it is problematic to tag exceptionally blurry
>     faces
>     as "ignore" in that this could result in some kind of "catch all
>     face"
>     that will match a lot of faces incorrectly?
>
>
>     PS. What about the "This is not a face" (red X button) - does this
>     also
>     just tag the detected face region as a "not face" tag? Or does it
>     rather
>     just throw the face out of the detected faces pile (thus,
>     resulting in
>     this image region being re-detected as a face some later time)? Or
>     how
>     does this work?
>
>     I have noticed that when I have a picture with high entropy
>     regions like
>     a heap of gravel or rocks, it will often detect faces in that
>     "noise". I
>     always click "This is not a face" but I have a feeling that the same
>     regions keep turning up when I do face detection again. But maybe
>     it is
>     just other regions of the same "noise"?
>
>
>     -- 
>     Mvh
>     Thomas
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.kde.org/pipermail/digikam-users/attachments/20230302/85cb82cb/attachment-0001.htm>