Issue with building a model - error obtained

[Debbie Lam]

Hi,

I encountered an error (see attached photo) while trying to use the code in Scalismo Tutorial 6: Building a shape model from data.

May I know what caused the error?

Thank you in advance!

[Marcel Luethi]

Hi Debbie

Unfortunately the error message is terrible (sorry for that), but I think what the error means is that the data you are using is not in correspondence. Being in correspondence implies that all the data have the same number of vertices and each vertex corresponds to the same semantic point on the surface. Could this be the problem.?

Best regards,

Marcel

--
You received this message because you are subscribed to the Google Groups "scalismo" group.
To unsubscribe from this group and stop receiving emails from it, send an email to scalismo+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/scalismo/424fca4a-4979-4a15-a2d9-bd5be7f776b5n%40googlegroups.com.

[Debbie Lam]

Hi, Marcel

I understand your point but my skull meshes were obtained from a database, in which the meshes were provided by different users. May I know for the PCA model building method using Data Collection, is correspondence a requirement? I highly doubt that the meshes that I've found have the same number of vertices.

Besides that, if my meshes don't share the same number of vertices, is there any method to solve it? Could they still be used to build a statistical shape model?

Thank you very much!

[Marcel Luethi]

Hi Debbie

Yes, you will first have to establish correspondence, i.e. register the data. This will be covered in the futurelearn course from around week 5 and you can find example code in the tutorials, in particular tutorial 12:

https://scalismo.org/docs/Tutorials/tutorial12

Best regards,

Marcel

To view this discussion on the web visit https://groups.google.com/d/msgid/scalismo/090a6d3e-14e2-4230-8b05-ccd34992a10fn%40googlegroups.com.

[Debbie Lam]

Hi Marcel

May I know besides the Parametric, non-rigid registration that you mentioned in Tutorial 12, is it possible for me to use the Iterative Closest Point (ICP) method to  establish correspondence?

[Marcel Luethi]

HI Debbie,

Yes, this is another possibility. The ICP approach is simple and very flexible. The main difference to the approach in Tutorial 12 is, that ICP does not need gradients to be calculated. This is, on one hand, an advantage, as it allows one to add constraints to improve the registration process. On the other hand, the gradient based approach does require less tuning to achieve good results.

Maybe you should just experiment with both and see which one gives better result for the particular shape that you would like to register.

Best regards,

Marcel

To view this discussion on the web visit https://groups.google.com/d/msgid/scalismo/9eeade77-db5d-4d93-98cc-f12610b02400n%40googlegroups.com.

[Debbie Lam]

Noted! I will try using the ICP approach!

Thank you very much :)