MaxPlanckSociety
Autonomous systems such as robots or self-driving cars are expected to play an increasingly important role in future. Initially, however, they must learn how to safely negotiate their environments. To this end, Michael Black teaches them to recognize people as such: a technology that will enable even more amazing applications.
Source
Intelligent Systems – computer learn to understand the world
When the kernels are applied in the first visual recognition steps ,is it favorable not to do perfectly shaped squares? but rather let the computer decide what the kernel shape and connection distribution is.
to maybe improve its performance.
Nice video, interesting work but the title is all wrong. Perception is not the same as understanding. If computer "understood" the world they would be able to come up with their own theories of how the world works. And notice it takes 60 fixed high resolution cameras in very controlled circumstances to do what two constantly moving eyes with very poor resolution and do in in a wide range of circumstances. I don't think we can be replaced …. yet.