Sciweavers

Share
CVPR
2011
IEEE

On Deep Generative Models with Applications to Recognition

8 years 2 months ago
On Deep Generative Models with Applications to Recognition
The most popular way to use probabilistic models in vision is first to extract some descriptors of small image patches or object parts using well-engineered features, and then to use statistical learning tools to model the dependencies among these features and eventual labels. Learning probabilistic models directly on the raw pixel values has proved to be much more difficult and is typically only used for regularizing discriminative methods. In this work, we use one of the best, pixel-level, generative models of natural images – a gated MRF – as the lowest level of a deep belief network (DBN) that has several hidden layers. We show that the resulting DBN is very good at coping with occlusion when predicting expression categories from face images, and it can produce features that perform comparably to SIFT descriptors for discriminating different types of scene. The generative ability of the model also makes it easy to see what information is captured and what is lost at each lev...
Marc', Aurelio Ranzato, Joshua Susskind, Volodymyr
Added 05 Apr 2011
Updated 29 Apr 2011
Type Journal
Year 2011
Where CVPR
Authors Marc', Aurelio Ranzato, Joshua Susskind, Volodymyr Mnih, Geoffrey Hinton
Comments (0)
books