I see that some new work flow is in the works for the faces work flow (GSoC 2020) . I have a few thoughts on what I would like to see.
My comments are based on how I understand things to work, so I could be off base.
1) Currently, if a face is detected, I have the option of tagging the face as "not a face", giving it a name, or just leaving it as it currently is. I would like to see an option that leaves the face, but ignores it as it is someone I really don't care about at this time (if ever) If faces tagged as not a face are used to re-train face detection, that is probably not a good thing.
2) If a face is recognised, but incorrect, I'd like to say the person is incorrect and feed that back into the training data.
3) If I manually add a face, it means that face detection didn't detect the face. It should be remembered that it was added manually and not be used to train. Sometimes I like to add "faces" into my photos that would make really poor data to train on (obscured faces, mostly turned away, etc)
4) Re-training the recognition database should also ignore previously manually added faces. I'd also like to see an option to re-train with just the initial database in case the currently detected faces represent as a whole poor candidates. I've been using Digikam for many years and I'll be honest, there is a lot of crap in my faces database. The ability to start re-training from scratch without throwing out my current faces seems like it would be a big help for future recognition attempts.
Rob
Locked
