CIS 680, Vision and Learning, Spring 2003

CIS 680-301, Vision and Learning, Spring 2004

Time & Place: TR 430pm-6:00pm, Levine 315
Instructor: Jianbo Shi, jshi@cis.upenn.edu

Course Description

In recent years, we have seen a convergence between machine vision and machine learning. The combination of the machine learning techniques with the right vision routines, have produced impressive results in object tracking, face detection, and object recognition. At the same time, successful though somewhat ad hoc vision algorithms have provided new insights to many machine learning problems.

In this course, we will explore this connection between vision and learning. We will cover topics in 1) image texture synthesis; 2) object detection and segmentation; 3) dynamic object tracking; 4) object and scene recognition; 5) human activity recognition and inference.

Project

The class project page is here.

Course Schedule

Date	Topics	Papers	Discussion
1/13	Texture: synthesis- a practical guide	Efros, Hertzmann, Shum	here
1/15	Texture: analysis-image statistics, similar measure	Martin & Fowlkes & Malik, Rubner & Tomasi & Guibas, Puzicha et.al.
1/20	Texture: synthesis/analysis: probabilistic formulation	Zhu & Wu & Mumford,
1/22	Texture: synthesis/analysis: probabilistic formulation	Portilla & Simoncelli
1/27	Object Detection: face detection- statistical approaches	Scheinderman & Kanade, Viola & Jones	Liming Zhao
1/29	Object Detection: more on boosting & bagging	Freund & Schapire breiman
2/3	Object Detection: flexible object detection via Graphical Models	Ioffe & Forsyth	Ankita Kumar
2/5	Object Detection: flexible object detection via Graphical Models	Felzenszwalb & Huttenlocher
2/10	Object Detection: efficient inference procedures for Graphical models(HMM, Tree, MRF):	Tutorial, Ghahramani & Jordan, Smyth &Heckerman,
2/12	Object Detection: Learning graphical models from examples	Song & Goncalves & Perona Fergus, Perona, & Zisserman
2/17	Object Detection: Review on EM, HMM	Bilmes
2/19	Object Detection: variational approach for graph inference	Jordan & Ghahramani & Jaakkola & Saul	Guest lecture by L. Saul
2/24	Object Tracking: Sampling, particle filtering	Isard & Blake Cham & Rehg
2/26	Object Tracking: Markov Chain Monte Carlo(MCMC) methods	Crisan & Doucet Tu & Zhu	Erdan Gu
3/2	Image Representation: PCA, ICA, Mixture Models	Bell & Sejnowski Roweis & Ghahramani	Hari Sundar
3/4	Image Representation: Learning Image Features	Lee & Seung Stauffer & Grimson
3/16	Object Recognition: Digit Recognition with Shape Context,	Belongie, Malik, Puzicha
3/18	Object Recognition: Digit/Face Recognition, Support Vector Machine(SVM),	Burges Vapnik,	Erdan Gu
3/23	Object Recognition: Neutral Net,	LeCun,	Hari Sundar
3/25	Object Recognition: Multi-class Object Recognition	Mahamud, Hebert and Lafferty	Fei Sha
3/30	Grouping: Object Segmentation: Graph cuts approaches	Shi, Malik, sharon, Brandt, Basri	Liming Zhao
4/1	Grouping: Stereophesis, Image labeling: Markov Random Field, and Graph Cuts	Ishikawa Geiger, Boykov, Veksler, Zabih	Ankita Kumar
4/6	Grouping: Grouping with Partial labeling	Yu & Shi
4/8	Grouping: Co-Training, knowledge transfer	Barnard, et. al., Blum & Mitchell,	Fei Sha
4/13	Grouping: Information bottleneck, clustering with side information	Tishby, Pereira & Bialek Hermes, Zoller and Buhmann, Peltonen, Sinkkonen and Kaski.
4/15	Action Recognition: Recognizing Human Movements	Bregler	Timothee Cour
4/20	Action Recognition: Learning Grammatical models of Human Actions	Moore & Essa
4/22	Review: Vision and Learning	Notes (20Mb)
	Action Recognition: Automatic Video Summarization
	Scene Recognition with Large Dataset
4/29	Project presentation

Course Format

This course consists of three components:

1) Course lectures. A tentative class schedule is listed above. A set of reference papers will be posted before each lecture.
On-line discussions on selected lecture topics. To encourage people to ask questions and get more out of each topic, an on-line discussion group will be created.Details of how to sign on can be found here. Each student is expected to lead one discussion topic. A written report is due one week after the lecture is given. I will email each of you thss projects. Projects could be done in a team of no more than two.

There is no class final examine. Project 50%, Selected topic discussion 40%, Class participation: 10%.

References

Vision Books:

Forsyth and Ponce, Computer Vision -- A Modern Approach
Trucco and Verri, Introductory Techniques for 3-D Computer Vision

Machine Learning Books:

Chris Bishop, Neural Networks for Pattern Recognition
Duda and Hart and Stork, pattern Classification

Matlab:

On-line Discussion: Quicktopic.com

last updated by jshi on January 16, 2004