Semantic Segmentation Exploiting Occlusion Relations within a Convex Optimization Framework

Semantic Segmentation Exploiting Occlusion Relations within a Convex Optimization Framework

Conference

Abstract

​We describe an approach to incorporate scene topology and semantics into pixel-level object detection and localization. Our method requires video to determine occlusion regions and thence local depth ordering, and any visual recognition scheme that provides a score at local image regions, for instance object detection probabilities. We set up a cost functional that incorporates occlusion cues induced by object boundaries, label consistency and recognition priors, and solve it using a convex optimization scheme. We show that our method improves localization accuracy of existing recognition approaches, or equivalently provides semantic labels to pixel-level localization and segmentation

 

Details

PUBLISHED IN
International Conference on Energy Minimization Methods in Computer Vision and Pattern Recognition (EMMCVPR)
PUBLICATION DATE
15 八月 2013
AUTHORS
B. Taylor, A. Ayvaci, A. Ravichandran, S. Soatto