Want to keep learning?

This content is taken from the University of Basel's online course, Statistical Shape Modelling: Computing the Human Anatomy. Join the course to learn more.

Scalismo Lab: shape modelling with Gaussian Processes and kernels

In this hands-on step, we will build upon the previous tutorial video and practice shape model creation from scratch, that is without any provided data.

We will start by creating a continous Gaussian Process based on a simple Gaussian kernel. We will then see how to tune the parameters of this kernel to obtain either more or less smooth or pronounced deformations.

We will also reproduce the Gaussian Processes created in the video, which notably lead to symmetric and spatially localised deformation fields, and check up on the fact that sample covariances are simply kernels learned from data.

In the full track version, we will focus on the differences between scalar-valued and matrix-valued kernels and learn how to implement our own kernels by extending the required Scalismo classes.

To access the tutorial document:

  1. Switch to Scalismo Lab.
  2. Select the Shape Modelling with Gaussian Processes and Kernels document under:
    Documents -> Fast track, for the fast track version
    Documents -> Full track, for the full track version

If you have questions, ask them in the comments section here on FutureLearn.

If you are doing the full track version, please remember that the exercises are optional. If you find them too hard, you can continue going through the tutorial without solving the exercises.

Did anything go wrong and you have a weird shape output? Post it in our Shapes Gone Wrong Padlet!

Share this article:

This article is from the free online course:

Statistical Shape Modelling: Computing the Human Anatomy

University of Basel