For those of us less technical, wouldn't any face covering impede facial recognition? Hat and muzzle leave just your eyes. Plus shades, i don't see how facial recognition could work.
It depends. If the AI was trained on maskless faces, then the coverings blocks the data it needs.
Add in masked and unmasked photos in the training set and the same AI would be able to learn to do with the missing data.
U/Baconofhope is mostly correct, except that these are separate systems. Like the vector distance between features is a distinct AI from one that looks for "features" of eyes, nose, mouth, etc.
Where AI IDing people from gait were made for drones that only see people from a top-down view.
You mean what we taught it, upload photos of your self yay. Now lets play upload photos of yourself as a child and we will show you what you look like when you're older(facebook did this to get your child photos to gage child + current age changes to train it). People fed the machine. Just like the captcha select photos with xyz or w/o xyz to show you arent a robot, we are teaching the borg.
Correct, and their rationale makes sense... because to train AI systems you need exponential amounts of data for it to learn, and why bother maintaining terrabytes of data in a database when you can create a format for people to send training data as a stream.
A common misconception with AI is that people only 'program' the 'structure' and then is just a matter of feeding it data until it starts generating accurate results. That's why AI often ends up racist, or has 'errors' like misgendering trannies.
For those of us less technical, wouldn't any face covering impede facial recognition? Hat and muzzle leave just your eyes. Plus shades, i don't see how facial recognition could work.
It depends. If the AI was trained on maskless faces, then the coverings blocks the data it needs.
Add in masked and unmasked photos in the training set and the same AI would be able to learn to do with the missing data.
U/Baconofhope is mostly correct, except that these are separate systems. Like the vector distance between features is a distinct AI from one that looks for "features" of eyes, nose, mouth, etc.
Where AI IDing people from gait were made for drones that only see people from a top-down view.
You mean what we taught it, upload photos of your self yay. Now lets play upload photos of yourself as a child and we will show you what you look like when you're older(facebook did this to get your child photos to gage child + current age changes to train it). People fed the machine. Just like the captcha select photos with xyz or w/o xyz to show you arent a robot, we are teaching the borg.
Correct, and their rationale makes sense... because to train AI systems you need exponential amounts of data for it to learn, and why bother maintaining terrabytes of data in a database when you can create a format for people to send training data as a stream.
A common misconception with AI is that people only 'program' the 'structure' and then is just a matter of feeding it data until it starts generating accurate results. That's why AI often ends up racist, or has 'errors' like misgendering trannies.