Date/Time
Date(s) - 11/07/2016
10:30 am - 11:30 am
Location
LSRC D344
Categories No Categories
Abstract: Shape retrieval is becoming an increasingly important problem because of the diffusion of 3D scanners and printers, and the creation of libraries of 3D shape models for a variety of applications. At the same time, deep neural nets have recently enabled significant progress in computer vision, particularly for object recognition from 2D images. I explore some of the literature at the nexus between these two areas, with the purpose of identifying interesting research directions. An important conceptual challenge is how to make the retrieval of 3D shapes invariant to similarities, that is, to translation, rotation, and scale, so that objects are recognized regardless of where they are viewed from. Will the mathematical sophistication of classical shape retrieval methods yield to the brute force of deep nets, or should the two approaches be harmonized?
