Probabilistic Morphable Models

Probabilistic Morphable Models are a fully probabilistic approach to model-based image analysis, based on:

• the theory of Gaussian processes for modelling shapes
• Markov chain Monte Carlo (MCMC) methods for model fitting
• a software framework for efficient development of image analysis solutions.

This course teaches the theoretical basis of shape modelling and model fitting and provides an introduction on how to use the software framework Scalismo. The participants have the chance to apply the learned concepts and methods in a guided real-world project from the area of computer vision or medical image analysis.

Course organization

The course consists of two parts: An open online course and a summer school.

In the online course, participants learn the basic theory of shape modelling and are given a practical introduction to Scalismo. Participants of this course, who submit a solution to the course project, are invited to attend the summer school.

The summer school will focus on model-based image analysis. In theory lectures, participants will learn more about model-fitting and registration, with particular emphasis on probabilistic model-fitting using Markov Chain Monte Carlo methods. An important part of the summer-school is a project. Depending on their preference and background, the course participants can choose to work on a real-world problem from the field of medical image analysis or face image analysis.

Target audience

The intended audience for the course is beginning graduate students and researchers in computer vision and medical image analysis, who plan to apply statistical shape models as part of their work.

Prerequisite is good knowledge of basic probability theory and statistics as well as linear algebra. Participants should have basic programming skills in at least one programming language (ideally Java, C++ or Python).