This is an old revision of the document!
Probabilistic Graphical Models in Computer Vision (IN2329) (2h + 2h, 5 ECTS)
Summary
Several problems in computer vision can be cast as a labeling problem. Typically, such problems arise from Markov Random Field (MRF) models, which provide an elegant framework of formulating various types of labeling problems in imaging.
By making use of certain assumptions some „nice“ MRF models can be solved in polynomial time, whereas others are NP hard. We will see both, efficient algorithms for solving the „nice“ problems and relaxation strategies for the „hard“ ones.
The following topics will be covered in this module:
Directed and undirected graphical models
- Bayesian Networks
- Markov Random Fields
- Conditional Random Fields
Parameter learning for MRF and CRF models
- Gradient based optimization
- Stochastic gradient descent
- Structured Support Vector Machines
Exact MAP inference methods for MRFs
- Belief propagation on trees: max-sum algorithm
- Binary graph cuts
- Branch-and-mincut
Approximate inference methods
- Loopy belief propagation
- Mean field approximation
- Graph cuts: alpha expansion, alpha-beta swap
- Linear programming relaxations: fast primal-dual schema
Practical applications that we will cover are:
- Binary and multi-label image segmentation
- Human pose estimation
- Stereo matching
- Object detection
Lecture
Location: Room 00.13.036
Time and Date: Tuesday 10:00 - 12:00
Lecturer: Dr. Csaba Domokos
Start: 12 April 2016
The lecture is held in English.
Tutorial
Location: Room 02.05.014
Time and Date: Tuesday 14:00 - 16:00
Organization: Lingni Ma
Start: 12 April 2016
Bonus
By handing in reasonable solutions to 60% of the exercises you can obtain a bonus of 0.3 in the final exam. Note that you can neither improve a 1.0 nor a 5.0.
Prerequisites
The course is intended for Master students.
The requirements for the class are knowledge in basic mathematics, in particular multivariate analysis and linear algebra, and in basic computer science, in particular dynamic programming and basic data structures.
Exam
The exam will be oral.
Recommended literature
- D. Koller, N. Friedman. Probabilistic Graphical Models: Principles and Techniques, MIT Press, 2009.
- S. Nowozin, C. H. Lampert. Structured Learning and Prediction in Computer Vision, Foundations and Trends in Computer Graphics and Vision, 2011. Download
- A. Blake, P. Kohli, C. Rother. Markov Random Fields for Vision and Image Processing, MIT Press, 2011.
Lecture Material
Course material (slides and exercise sheets) can be accessed here.
The password will be presented in the first lecture.