Pretty cool algorithm, I could see this being used in lower cost Mo-Cap applications or 3D scanning without expensive hardware.
Holy shit, in a few years we surpassed just being able to train digital neural networks to recognize what they see, but also how to infer a 3D structure based on its own underlying assumptions of how it should look in a prototypical fashion - like how we do.
I'm assuming this uses a lot of shortcuts in order to optimize it to strictly the shape of the human body, but if this can be abstracted out to accept any geometrical shape as the comparative/prototype input, then that system would essentially emulate a low level mammalian occipital lobe, specifically the ventral portion.
So I know I already said I wanted the physics-simulated sounds, but...
Want.
Also, am I a bad person if I first thought this video was being narrated by [b][url=https://www.youtube.com/watch?v=f5VQ_c5v4XM]Kilian Experience[/url][/b]?
araki will love this
[QUOTE=Karmah;51381730]Holy shit, in a few years we surpassed just being able to train digital neural networks to recognize what they see, but also how to infer a 3D structure based on its own underlying assumptions of how it should look in a prototypical fashion - like how we do.
I'm assuming this uses a lot of shortcuts in order to optimize it to strictly the shape of the human body, but if this can be abstracted out to accept any geometrical shape as the comparative/prototype input, then that system would essentially emulate a low level mammalian occipital lobe, specifically the ventral portion.[/QUOTE]
I'm not too sure on that. It uses a convolutional neural network, which is a massive over-simplification to what animals do in their brains.
Sorry, you need to Log In to post a reply to this thread.
