[digiKam-users] Searching unconfirmed face recognition results

[_]

Hi all, I have a large collection (~200k) of photos, and I'd like to filter
it down to a certain set of people/faces with a minimal amount of manual
work.

The face recognition tool is very helpful for this, but I've run into some
roadblocks: Most faces are correctly recognized, but not tagged until I
confirm the result. I can't search for people without confirming and adding
the face tags, yet I don't want to have to confirm all the matches because
it's infeasible for me to check hundreds of thousands of faces. I don't
want to blanket confirm everything, because not all results are correct,
and I suspect adding many low quality and incorrect matches will screw up
the recognition algorithm.

Ideally, I think the tool should mark all recognized faces with an
"unconfirmed" subtag. So if "Emily" is a person, then the face recognition
should tag the matches as "People/Emily/unconfirmed", and the user can (but
does not have to) confirm and move them into the "People/Emily" tag.
Importantly, the face recognizer should only use the confirmed images as
training/matching data. I've already manually implemented these sub-tags,
but I don't have any way of getting the tool to ignore sub-tagged images in
the recognition process (it's not clear from the documentation how the
recognizer handles nested People tags, but it seems to treat them as
different people? I might be mistaken here). Also, if I want to add new
images and rebuild the training data, I would have to delete and recreate
all my subtags, which is a real pain.

Curious if anyone else has a better workflow for doing this

Thanks,
Shimao
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.kde.org/pipermail/digikam-users/attachments/20211010/7e8aaef7/attachment.htm>