https://bugs.kde.org/show_bug.cgi?id=432207

Thilo <[email protected]> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |[email protected]

--- Comment #2 from Thilo <[email protected]> ---
(In reply to Maik Qualmann from comment #1)
> It does not work like that. We don't have a next match. You have to assign
> more people first for the face engine to learn. This is not a fingerprint
> comparison.
> 
> Maik

Hello, I think this is an interesting question. Labelling, that a face does not
belong to a certain person is valuable information. I understand, that it is
not possible to simply reject this classification result and to accept the next
best match (which would work when there would be something like a "similarity"
vector for a query face that gives a scalar value of similarity for every
person in the database.  
However, wouldnt it be possible to use the end users rejection of a
classification result as "negative labeling" information for a retraining,
leading to an improved classification result?

-- 
You are receiving this mail because:
You are watching all bug changes.

Reply via email to