RE: Asking for your help about statismo's femur example

[Marcel Luethi]

(This e-mail is in response to a question asked on a separate thread and github issues, see below)

Dear Zhiqi Zhao,

The process of preparing the meshes, such that all the meshes have the same points is called “establishing correspondence” and is done doing non-rigid registration.
You find an example of how to do this in statismo on the wiki:
https://github.com/statismo/statismo/wiki/cli-femur-model-example

The landmarks we are  using for alignment and registration are defined manually. If you would like to read more about shape modelling and registration, we offer an online couse on shape modelling. You can access it here:
http://shapemodelling.cs.unibas.ch/StatisticalShapeModelling_2018.html
The course uses Scalismo (a more modern version of Statismo, using Scala), but the concepts equally apply to statismo.

 Best regards,

 Marcel

Dear Mr. luethi:

         My name is Zhiqi Zhao. I am a Master graduate student in Beijing Institute of Technology, China. I  have read your paper titled<Gaussian Process Morphable Models> and build your framework named "Statismo".So i have some questions and i want to ask your help! In your given femur example , How do you keep the points number of all femur meshes consistent? they are all 171380 points, or what method did you use to generate femur mesh data? For your landmark constraints, how do you choose the corresponding landmarks? manually or other methods? I'm looking forward to your reply!

[zhaozh...@gmail.com]

That is to say, in https://github.com/statismo/statismo/wiki/cli-femur-model-example , before using the GP model to do the registration to other training instances, the non-rigid registration of the training samples has been done?

在 2018年8月7日星期二 UTC+8下午1:53:43，Marcel Luethi写道：

[Marcel Luethi]

There are two different GP Models involved in the process. The first, analytically defined model, is needed to do the registration. A second GP model (which is a traditional Statistical Shape Model), is learned from the registered example surfaces.

Best regards,

Marcel

--
You received this message because you are subscribed to the Google Groups "statismo" group.
To unsubscribe from this group and stop receiving emails from it, send an email to statismo-user...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[zhaozh...@gmail.com]

According to your reply, I think what I am talking about is the first GP model, which is defined by a reference training instance.  What I want to ask is, when using this GP model to register to other training samples, should i need to keep the number of points of the meshes consistent, that is, must i first perform non-rigid registration?

在 2018年8月7日星期二 UTC+8下午2:57:11，Marcel Luethi写道：

[Marcel Luethi]

The GPModel is used for establishing correspondence (i.e. non-rigid registration). The result of this process are meshes, which all have the same number of vertices. But for performing the registration, the meshes do not need to have the same number of points.

I would advice you to have a careful look at the example

https://github.com/statismo/statismo/wiki/cli-femur-model-example

and to also take a look at the online course, if the concepts confuse you.

Best regards,

Marcel

[zhaozh...@gmail.com]

      Thanks for your reply! I understand what you mean. After reading and running the example of the femur you gave, I found that the number of points you used for registration was the same before registration. I tried to register to my data (there are different points, but you said it doesn't matter with the number of points), and found that the registration result is not good. I know that when using the GP model to register to your given data, the effect is not very good too. In order to better register the results, you manually added the corresponding point constraints. After adding the landmarks constraint, I found that your registration result is very good! 

     So I have some questions I would like to ask you: How do you determine the coordinates of corresponding landmarks manually added? Select anatomical landmarks through experience? Still have some clever ways? I tried to apply data different from the bones (such as the optic nerve), is it theoretically feasible? Is the parameters (-s, -n, -p) for establishing the GP model adjusted according to the actual situation?

     Looking forward to your reply!

在 2018年8月7日星期二 UTC+8下午4:57:44，Marcel Luethi写道：