Unsupervised Learning Predicts Perception and Misperception of Materials

Unsupervised Learning Predicts Perception and Misperception of Materials

Abstract

Presenting new results from two projects using unsupervised deep neural networks to learn about image structure from static or moving visual input. Such networks spontaneously learn to represent underlying scene factors, and can predict human gloss perception and misperception on an image-by-image basis.

Date
Event
Symposium Talk
Location
SFB Symposium: How Humans and Machines Learn to See, Rauischholzhausen Castle, Germany