Correcting face recognition

[jln]

Face recognition is useful when importing photos, and I have it set to create keywords so I can search for people.

What I am unsure about is the benefit of correcting face recognition when it either misidentifies a face or fails to recognize a face.

Is there a benefit to correcting face recognition rather than just adding a keyword with the person's name to the image? Does it improve face recognition? Face recognition has difficulty (understandably) identifying faces in profile. Is there a benefit to labeling such a face?

I think part of my question has to do with training. Does training only occur once when you select the faces to train on, or does it update when errors are corrected? I understand the AI doesn't work if the training set is too large, but isn't the optimal size of the training set partially a function of the total number of faces it is called upon to identify. Should the training set ever be changed or added to?

[Mario]

Running face recognition, creating persons, assigning persons to faces etc. is entirely up to you. IMatch will work and organize your files with or without face recognition and persons.

Correcting a wrongly assigned person (or manually adding a face and person when IMatch did not detect a face itself) adds
a) the information that the image contains a face a where
b) the person the face belongs to
c) the keywords associated with the person (if you use that feature)
d) the categories associated with the person (if you use that feature)

Knowing if an image contains one or more faces allows IMatch to provide features based on Face Arrangement
Knowing the persons in an image allows IMatch to search for the persons, organize images by the persons they contain, associate persons with events etc.

A wrong person assigned to a face will cause IMatch to miss the file when you search for the "correct" person, the wrong keywords will be assigned etc. The idea of having keywords linked to a person is that IMatch automatically adds (or removes) the keywords when you assign a person to a face or change the person when IMatch assigned it wrong.

Correcting a wrongly assigned person or adding a face where IMatch missed it and assigning a person is a "once only" task - but the benefits like search, organize etc. will stay forever.

I generally correct mistakes made by IMatch. I don't shoot thousands of images per week, though!
Reviewing a few hundred images and correcting the occasional missing face or wrong person assignment is a matter of a couple of minutes and well worth it IMHO.

Changing the trained faces for a person may be a good idea if you have images of the person at a wide variety of ages. IMatch automatically trains the first 10 faces you assign to a person - which may be fine or could be improved by swapping too similar faces with some with more variety.

Training too much can be bad (I have seen databases where users trained 50 to 100 faces per person). 5 to 10 trained faces is good, unless you have images of the person at ages 10 to 90, where some more trained faces will do good.