مقاله On Learning Conditional Random Fields for Stereo : تی پی بین

خلاصه مقاله

Until recently, the lack of ground truth data has hindered the application of discriminative structured prediction techniques to the stereo problem. In this paper we use ground truth data sets that we have recently constructed to explore different model structures and parameter learning techniques. To estimate parameters in Markov random fields (MRFs) via maximum likelihood one usually needs to perform approximate probabilistic inference. Conditional random fields (CRFs) are discriminative versions of traditional MRFs. We explore a number of novel CRF model structures including a CRF for stereo matching with an explicit occlusion model. CRFs require expensive inference steps for each iteration of optimization and inference is particularly slow when there are many discrete states. We explore belief propagation, variational message passing and graph cuts as inference methods during learning and compare with learning via pseudolikelihood. To accelerate approximate inference we have developed a new method called sparse variational message passing which can reduce inference time by an order of magnitude with negligible loss in quality. Learning using sparse variational message passing improves upon previous approaches using graph cuts and allows efficient learning over large data sets when energy functions violate the constraints imposed by graph cuts.

نویسندگان

Christopher J. Pal [ École Polytechnique de Montréal, Montréal, QC, Canada ]

Jerod J. Weinman [ Dept. of Computer Science, Grinnell College, Grinnell, IA, USA ]

Lam C. Tran [ Dept. of Electrical and Computer Engineering, University of California San Diego, San Diego, CA, USA ]

Daniel Scharstein [ Middlebury College, Middlebury, VT, USA ]

کلمات کلیدی

Stereo Learning Structured prediction Approximate inference

آدرس این مقاله

آدرس مستقیم دسترسی به این مقاله

مشخصات کنفرانس

مشاهده مقالات

International Journal of Computer Vision 2013

International Journal of Computer Vision (IJCV) details the science and engineering of this rapidly growing field. Regular articles present major technical advances of broad general interest. Short articles provide a fast publication path for novel research results. Survey articles offer critical reviews of the state of the art and/or tutorial presentations of pertinent topics. Coverage includes: - Mathematical, physical and computational aspects of computer vision: image formation, processing, analysis, and interpretation; machine learning techniques; statistical approaches; sensors. - Applications: image-based rendering, computer graphics, robotics, photo interpretation, image retrieval, video analysis and annotation, multi-media, and more. - Connections with human perception: computational and architectural aspects of human vision. The journal also features book reviews, position papers, editorials by leading scientific figures, as well as additional on-line material, such as still images, video sequences, data sets, and software.

حوزه های تحت پوشش : IT & Software, Electronics, Engineering, Aerospace, Telecommunications, Automotive