That extent I'm not completely certain, I understand that in its basic sense, AI is a map of "neurons" with connections to other layers of neurons. The configuration and mapping can create different effects.
Really think about seeing someone you know in a mask, yes, eyes are key and likely does reflect in the "weights", head shape, nose bridge, eyes, lips, cheeks... the better you know someone the less of their face you would need ro see to recognize. (In this context there's a difference between X is a face and X face belongs to Y). It's that kind of pattern recognition that I think AI could pick up on.
People have done visualizations of what "motivated" the ai to one decision over another, and I have no reason to doubt that eye details have a stronger weighting, they are something of a black box where you give it input and get an output without really knowing how.
Ex; one test of an AI system found that, as far as the AI cared, the difference between a wolf and a German Shephard was.... wait for it... whether or not there was snow in the background. A wolf in the summer is a German shepherd.
It's a topic I took an interest in for a time, recreated a few "hello world" projects until i got to a point where my available hardware could not handle the training.
I kinda wanted a feel for how it worked to sort out the reality from what movies show.
Yes and no...
No in that an AI trained on maskless faces would lose accuracy.
Yes in that it wouldn't even necessarily need to be remade from scratch to learn a face with the missing data covered by a mask.
That extent I'm not completely certain, I understand that in its basic sense, AI is a map of "neurons" with connections to other layers of neurons. The configuration and mapping can create different effects.
Really think about seeing someone you know in a mask, yes, eyes are key and likely does reflect in the "weights", head shape, nose bridge, eyes, lips, cheeks... the better you know someone the less of their face you would need ro see to recognize. (In this context there's a difference between X is a face and X face belongs to Y). It's that kind of pattern recognition that I think AI could pick up on.
People have done visualizations of what "motivated" the ai to one decision over another, and I have no reason to doubt that eye details have a stronger weighting, they are something of a black box where you give it input and get an output without really knowing how.
Ex; one test of an AI system found that, as far as the AI cared, the difference between a wolf and a German Shephard was.... wait for it... whether or not there was snow in the background. A wolf in the summer is a German shepherd.
It's a topic I took an interest in for a time, recreated a few "hello world" projects until i got to a point where my available hardware could not handle the training.
I kinda wanted a feel for how it worked to sort out the reality from what movies show.