Bringing Traditional Computer Vision into Deep Learning

Mathieu Salzmann (EPFL Lausannne, Switzerland)

Abstract:

In this talk, I will discuss our recent progress in developing algorithms that incorporate traditional computer vision concepts into deep learning based approaches. In particular, in the first part of the talk, I will introduce a new type of deep architecture that exploits the notion of covariance descriptors employed in the past for visual recognition. I will show that this architecture can further be motivated from a statistical point of view. In the second part of the talk, I will describe an approach to performing novel view synthesis that exploits notions of geometry within a deep learning framework.