Training the AI facial recognition

mastersian

I have a large collection of photos all of which have now been scanned. I’ve chosen 5 faces to recognise, 3 of them are coming through at pave but two of the faces are hardly being recognised. Is there way to help “train” the AI to make it more effective at finding these faces?

Andrey

mastersian if I understand right, you have added several people to the database, but application does not offer more faces for some of those people for confirmation, but you are sure they are present in your archive. In that case I would recommend just give more samples to AI to train on. You can manually mark more photos of those people. That will give AI more food for thought.

mastersian

that’s what i thought, just wanted to confirm, thank you

Andrey

There is another feature you should be aware of. AI provides suggestions in small batches, and it prefers to suggest only those that have higher probability of being correct. So, if for example it sees lots of “low hanging fruits” for just one of people, it may focus on suggesting those first. Then, after you confirm those and all “low hanging fruits” will be covered, it may attend to other people too. The goal is to simplify it for user, by ensuring user mostly confirms, instead of cherry-picking correct suggestions among false ones, which could be tiresome. The flip side is, that suggestions are not always evenly distributed among people in your gallery.