Bayesian Nonparametric Methods: Hope or Hype?

Spatial Bayesian Nonparametrics for Natural Image Segmentation

author: Erik Sudderth, Brown University
published: Jan. 24, 2012, recorded: December 2011, views: 375

Slides

Slides
0:00	Spatial Bayesian Nonparametrics for Natural Image Segmentation
0:05	Parsing Visual Scenes
0:25	Region Classification with Markov Field Aspect Models
1:41	Human Image Segmentation
2:16	Berkeley Segmentation Database & Boundary Detection Benchmark
2:31	BNP Image Segmentation
3:33	The Infinite Hype
4:15	Some Hope: BNP Segmentation
4:39	Pitman-Yor Processes
5:22	Pitman-Yor Stick-Breaking
5:48	Human Image Segmentations
6:15	Statistics of Human Segments
7:11	Why Pitman-Yor?
8:28	An Aside: Toy Dataset Bias
9:19	Feature Extraction
9:53	Pitman-Yor Mixture Model
10:02	Dependent DP&PY Mixtures
12:09	Example: Logistic of Gaussians
12:44	Dependent DP&PY Mixtures
12:57	Discrete Markov Random Fields
13:26	Phase Transitions in Action
14:03	Product of Potts and DP?
14:04	Spatially Dependent Pitman-Yor - 1
14:57	Spatially Dependent Pitman-Yor - 2
15:07	Spatially Dependent Pitman-Yor - 3
15:33	Spatially Dependent Pitman-Yor - 4
16:17	Samples from PY Spatial Prior
17:00	Outline - 1
17:49	Mean Field for Dependent PY
19:13	Robustness and Initialization
19:50	Alternative: Inference by Search
21:30	Discrete Search Moves
21:54	Inference Across Initializations
23:11	BSDS: Spatial PY Inference
23:37	Outline - 2
23:40	Covariance Kernels
24:40	Learning from Human Segments
25:24	From Probability to Correlation
26:24	Low-Rank Covariance Projection
27:09	Prediction of Test Partitions
27:50	Comparing Spatial PY Models
27:52	Outline - 3
27:56	Other Segmentation Methods
29:06	Quantitative Comparisons
30:12	Multiple Spatial PY Modes - 1
30:40	Multiple Spatial PY Modes - 2
31:04	Spatial PY Segmentations
31:24	Conclusions

Report a problem or upload files

If you have found a problem with this lecture or would like to send us extra material, articles, exercises, etc., please use our ticket system to describe your request and upload the data.
Enter your e-mail into the 'Cc' field, and we will keep you updated with your request's status.

Lecture popularity: You need to login to cast your vote.