Face recognition

[hluxem]

Hello Mario,

I didn't want to convolute a feature request with some more general questions. This was stated in the batch confirming feature request:

My gut feel, from the
testing I did so far is that there is something else wrong which produces a lot of miss-matches. Many of the miss-matches
I encounter seem so wrong that I came to this conclusion

The limited testing I did with Imatch looked good and I thought the suggestions were mostly spot on.
I have used LR before and think it got about 80% right, you stated something like 95% in general. I do think the hits in Imatch are much better than in LR for sure. To be fair the version I have for LR is 5 years old.
One thing I saw was that some times family members got tagged wrong. Like my brother was tagged as my dad. That makes sense to me as family members look similar.

As I have about 600 people I'm wondering what is the impact of the number of people in the database. Does the number of people in the database have an impact on the accuracy of the suggestion? With LR I always started a new catalog and started from scratch for a batch of images. Therefore there were never that many different people.
If you have pictures of your kids, is the AI impacted when creating a fingerprint for a face from a mix of baby and teenager pictures?

Thanks,

Heiner

[Mario]

IMatch learns a number of faces for each person.
When a person needs to be assigned to a face, it compares learned faces of all persons in the database to find the best match.

I have test databases with up to 200,000 files and 1000 different persons. The recognition rate is very good.

[Carlo Didier]

The few false recognitions I have had so far were between close family members (siblings, mother and daughter) which can easily be understood by the logical similarities of the faces.
Worst case would of course be twins ...