News for hackers. Stuff that matters.

A comment on Why do CNNs generalize so poorly to small image transformations?.

Re:Why do CNNs generalize so poorly to small image transform... by d--b on Monday June 11, 2018 @05:44AM (#17283009)
Perhaps, I'm saying just perhaps, there is a reason why humans are good at discerning things in the very narrow field of vision that's straight ahead, and not very good at the peripheral vision.

Pointing first and then classify may help in solving those issues. Maybe?

[ Parent | on HN ]
Re:Why do CNNs generalize so poorly to small image transform... by PeterisP on Monday June 11, 2018 @07:35AM (#17283417)
The reason why humans are good at discerning things in the very narrow field of vision that's straight ahead, and not very good at the peripheral vision is biological, there's simply a much lower density of receptors ("pixels") in the periphery, so in the periphery there's much less sensory information for the brain to work with.
[ Parent | on HN ]