《Label Transfer via Dense Scene Alignment》.pdfVIP

Nonparametric Scene Parsing: Label Transfer via Dense Scene Alignment Ce Liu Jenny Yuen Antonio Torralba Computer Science and Artificial Intelligence Laboratory, Massachusetts Institute of Technology {celiu,jenny,torralba}@csail.mit.edu Abstract In this paper we propose a novel nonparametric ap- proach for object recognition and scene parsing using dense scene alignment. Given an input image, we retrieve its best matches from a large database with annotated images us- ing our modified, coarse-to-fine SIFT flow algorithm that aligns the structures within two images. Based on the dense (d) (e) window scene correspondence obtained from the SIFT flow, our sys- tree (a) sky tem warps the existing annotations, and integrates multi- road ple cues in a Markov random field framework to segment pole car and recognize the query image. Promising experimental re- building