Image understanding at a large-scale: From shallow to deep and beyond
Jorge Sanchez
National Scientific and Technical Research Council, Argentina
: J Comput Eng Inf Technol
Abstract
The analysis and understanding of images at a large-scale is a problem which has received an increasing attention in the past few years. The rapid growth on the number of images and videos online and the availability of datasets consisting on hundreds of thousands or even millions of manually annotated images, impose exciting new challenges to the computer vision community as a whole. One of the fundamental problems on visual recognition, i.e., the way we represent the images and its content, is witnessing a paradigm shift towards a new class of models trying to exploit the vast amount of available data as well as the fast growth and widespread use of high-performance computing systems. In this talk, I will discuss different models that have been proposed in the computer vision literature to encode the visual information over the past few years, from the early shallow models to the more recent deep architectures. I will focus on the large-scale image annotation problem, i.e., the task of providing semantic labels to images when the number of those images and/or the number of possible annotations is very large, and its connection with other problems of growing practical interest.
Biography
