Quiz

Learning shape variation from data is the key to characterising shape families. In principle, this is a straightforward process. With this quiz we would like to make you aware of some of the problems and subtle issues that can occur in practice.

Question 1

Assume we are given perfect correspondence for a set of shapes. Does this always guarantee that we obtain a good shape model? Multiple answers are possible.

Yes, correspondence always gives a good estimation of the shape variation.

Not if we still have rotations or translations in the model.

Not if the shape examples are not representative.

No, we could still compare the wrong anatomical parts.

Question 2

Assume you have a set of hands and you align all of them such that the tip of the thumb is superimposed in all the shapes. What can you say about the resulting model?

There is shape variation everywhere.

There is no shape variation at the point where the tip of the thumb lies.

The largest variance is at the tip of the thumb.

Question 3

Assume you have shapes in correspondence that are perfectly aligned (no variations due to rotation or translation). Will the shape variations you estimate differ depending on which reference shape you choose (i.e. from which shape you compute the deformations)?

Yes, the mean deformation changes and hence the modelled shapes are different.

No, the modelled shape variations will always be the same.

Question 4

Which of the following statements about the number of example shapes are true? Multiple answers are possible.

The more examples we have, the more accurate are our estimations of the sample mean and sample covariance.

If we have too few example shapes, we cannot explain the full shape variation within a shape family.

The learned shape variation always increases, independently of how many shapes we observe.