Quiz

We have seen that classical shape models are not always able to represent all the shapes of a given shape family. This quiz gives you the opportunity to check if you understand why this is the case and deepen your knowledge about possible remedies for this problem.

Question 1

What problems can occur if we do not have a sufficient number of example datasets from which we can learn the deformations? Multiple answers are possible.

We cannot estimate the mean deformation accurately.

We cannot estimate the covariance function accurately.

We may not see some of the shape variations in the example data and hence these will not be represented in the model.

Question 2

What happens to the modelled variance at a certain point of the shape when we add two covariance functions together?

The variances sum up and hence there is more shape variation at this point.

The variances are not influenced by this and the shape variation remains the same.

Question 3

We have seen that we can enlarge the flexibility of a statistical shape model by adding a Gaussian kernel with a large standard deviation σ. What assumption do we make in this model?

The part that we cannot explain with the learned shape variation can be well explained by a smoothly varying deformation, ie the deformations are all highly correlated.

The part that we cannot explain with the learned shape variations has a large variance.

The part that we cannot explain with the learned shape variation can be modelled using uncorrelated, Gaussian noise.

Question 4

Which of the following measures would also increase the flexibility of a statistical shape model? Multiple answers are possible.

Multiplying the covariance function by a positive scalar > 1.

Subtracting a Gaussian kernel.

Multiplication with a Gaussian kernel.