vwlsmssng 3 points 2y ago
Probably by using neural networks.
https://link.springer.com/article/10.1007/s10462-019-09765-w
Rethunker 2 points 2y ago
When you train a neural network to distinguish between objects, you feed the network a lot of images that are labeled. Here “labeled” means some description is written.
Notice that in Seeing AI that when you switch to the mode to recognize people, it is no longer reading text or recognizing objects. Those require different training.
RJHand 2 points 2y ago
No idea, ask microsoft. Probably has something to do with the algorithm (s) in place. Its also worth noting though that even sighted people probably struggle with that too, you can't guess accurately every time. Seeingai may be a bit further off than visually guessing, probably is based on my experiments with it, but its also a bit more realistic too as we also have to guess about things too same as anyone else would seeing said photos lol.