Fitting parametric models to novel data is a common problem in many disciplines. In Computer Vision and Medical Image Analysis such optimization tasks are often difficult due to the large number of model parameter and additionally the problem of local minima of the fitting problem.
In this free online course, you will get insights how probabilistic methods can overcome these difficulties. You will learn how the optimization problem can be reformulated in a fully probabilistic form using "Bayes-Theory". Based on this formalism you will understand that a data-driven Markov Chain Monte Carlo optimization technique is well suited for the problem.
During the tutorial you will implement a simple framework to reconstruct a face from a single photograph.
The structure of this tutorial is similar to the Statistical Shape Modelling course, it consists of:
- interactive software tutorial.
All course material is wrapped into single tutorial file that must be downloaded as a whole:
Download Probabilistic Fitting Tutorial
- Modern CPU (two cores to keep the GUI responsive)
- Screen size > 800x600
- RAM 2GB
- Recent Java 8 JVM, > 1.8.0_80 (Oracle and OpenJDK should work) (use 64 bit version if available)
- Download the Basel Face Model 2017 and add it to the data directory (model2017-1_face12_nomouth.h5).
- Make sure you start the tutorial in the working directory where the data folder is contained.
To prevent memory issues, you can launch the tutorial using the extra JVM flag -Xmx with an explicitly set amount of memory. Use 2g for optimal results.
Start the tutorial with the command:
java -Xmx2g -jar scalismoLab-faces.jar