Event Information:
Erik Sudderth
(Assistant Professor, Dept of Computer Science, Brown University)
Shared Segmentation of Natural Scenes using Dependent Pitman-Yor Processes
Statistical Science Seminar
Shared Segmentation of Natural Scenes using Dependent Pitman-Yor Processes
- Abstract:
- We explore statistical frameworks for the simultaneous, unsupervised segmentation and discovery of visual object categories from image databases. Examining a large set of manually segmented scenes, we show that object frequencies and segment sizes both follow power law distributions, which are well modeled by the Pitman-Yor (PY) process. This generalization of the Dirichlet process leads to learning algorithms which discover an unknown set of objects, and segmentation methods which automatically adapt their resolution to each image. Generalizing previous applications of PY priors, we use non-Markov Gaussian processes to infer spatially contiguous segments which respect image boundaries. Using a novel family of variational approximations, our approach produces segmentations which compare favorably to state-of-the-art methods, while simultaneously discovering categories shared among natural scenes. Moreover, our hierarchical Bayesian approach provides a convenient framework for transferring human annotations among different image collections.
Reception immediately following in 211 Old Chemistry
Statistical Science Seminar