About The Forum
Calendar 2009 Poster Sessions : Decomposing a Scene into Geometric and Semantically Consistent Regions
Abstract:
High-level, or holistic, scene understanding involves reasoning about objects, regions, and the 3D relationships between them. This requires a representation above the level of pixels that can be endowed with high-level attributes such as class of object/region, its orientation, and (rough 3D) location within the scene. Towards this goal, we propose a region-based model which combines appearance and scene geometry to automatically decompose a scene into semantically meaningful regions. Our model is defined in terms of a unified energy function over scene appearance and structure. We show how this energy function can be learned from data and present an efficient inference technique that makes use of multiple over-segmentations of the image to propose moves in th energy-space. We show, experimentally, that our method achieves state-of-the-art performance on the tasks of both multi-class image segmentation and geometric reasoning. Finally, by understanding both region classes and geometry, we show how to generate 3D reconstructions of the scene which compete with more complex state-of-the-art methods.
Bio:
Stephen is a PhD student at Stanford University studying machine learning and artificial intelligence. His primary research focus is the use of probabilistic models in computer vision. Stephen's PhD advisor is Daphne Koller. He also works frequently with Andrew Ng.
Prior to resuming studies for his PhD, Stephen co-founded Sensory Networks, a network security company, where as VP Engineering, he took the company from its inception to over 80 employees. Before Sensory Networks, Stephen worked in various engineering positions in research and development companies (including Dilithium Networks, Polartechnics, and CWC). Stephen received his MSEE degree from Stanford in 1998, and BE and BSc degrees from The University of Sydney, Australia in 1996 and 1994. Stephen holds eight international patents and is the author on a number of technical papers. He enjoys mountain bike riding, playing golf, some traveling, and solving puzzles.
High-level, or holistic, scene understanding involves reasoning about objects, regions, and the 3D relationships between them. This requires a representation above the level of pixels that can be endowed with high-level attributes such as class of object/region, its orientation, and (rough 3D) location within the scene. Towards this goal, we propose a region-based model which combines appearance and scene geometry to automatically decompose a scene into semantically meaningful regions. Our model is defined in terms of a unified energy function over scene appearance and structure. We show how this energy function can be learned from data and present an efficient inference technique that makes use of multiple over-segmentations of the image to propose moves in th energy-space. We show, experimentally, that our method achieves state-of-the-art performance on the tasks of both multi-class image segmentation and geometric reasoning. Finally, by understanding both region classes and geometry, we show how to generate 3D reconstructions of the scene which compete with more complex state-of-the-art methods.
Bio:
Stephen is a PhD student at Stanford University studying machine learning and artificial intelligence. His primary research focus is the use of probabilistic models in computer vision. Stephen's PhD advisor is Daphne Koller. He also works frequently with Andrew Ng.
Prior to resuming studies for his PhD, Stephen co-founded Sensory Networks, a network security company, where as VP Engineering, he took the company from its inception to over 80 employees. Before Sensory Networks, Stephen worked in various engineering positions in research and development companies (including Dilithium Networks, Polartechnics, and CWC). Stephen received his MSEE degree from Stanford in 1998, and BE and BSc degrees from The University of Sydney, Australia in 1996 and 1994. Stephen holds eight international patents and is the author on a number of technical papers. He enjoys mountain bike riding, playing golf, some traveling, and solving puzzles.