Open luzik opened 2 years ago
You can upload face examples that do not face forward. It should improve the accuracy. You can try to calculate the angle using the landmarks plugin.
I have a few samples trained, but its still gives false positive.
Please look at todays photo (I am not "amelka")
Does double take
show the probability that this is a face? Because this is a problem, the system shouldn't even try to recognize this face. I mean it should be sean at least half of the face
I have some false positives when person do not facing forward(even 99% hit), especially if it facing about 90deg to the left or right (side face). Can I prevent face recognition if subject is not facing forward ? Is there any information about facing angle ? Maybe mask provide such informations ?