Superpixel Graphs for Semantic Scene Understanding
Stephen Gould (ANU)
NICTA SML SEMINARDATE: 2013-10-17
TIME: 11:15:00 - 12:15:00
LOCATION: NICTA - 7 London Circuit
CONTACT: JavaScript must be enabled to display this email address.
ABSTRACT:
Semantic segmentation is a fundamental step for automatically understanding images. In this task every pixel in an image is annotated with a category label such as "sky", "tree", "water", etc. Traditional methods learn a per-pixel classifier for each category of interest and combine these with pairwise and higher-order constraints via a conditional Markov random field. The difficulty with such methods is that they scale poorly with the number of images and the number of categories. Recent work on label transfer methods aim to address the problem of growing data-sets and changing categories. In this talk I will present our work on a graph-based nearest neighbor method for scene understanding with learned distance metrics.
BIO:
http://users.cecs.anu.edu.au/~sgould/
