As I am testing Dlib a little deeper, I am seeing a lot of false positive especially at night in spite of having set the confidence to 0.3.
What does this setting mean and what range can it be set to?
I have setup two instances with two faces and noticed that when one gets a false positive, the other does as well. The real face recognition only occurs most of the time when only one of the two is recognized. Anybody knows?
Answering my own question after digging into it and doing quite a bit of research and optimization:
This parameter actually is used a comparator from the face_recognition package which serves as a criterion for the distance between two faces calculate from the 128 point using NumPy.
Very long way to say that it is model dépendant and an arbitrary number which needs to be tested according to the face database…