No selected corresponding points on the nose

[Maia R.]

Hi Dear Marcel et community,

I am working with Tutorial 11 on non rigid ICP for model fitting.

I don't understand why I can't have more points on the nose model (see figure).

Also, I added more and more points, starting from 5000 to  20000,  and also increased the number of iterations (20 to 100).  I observed that the quality of the fit by computing the metrics  (e.g., avgDistance) does not change and had no dramatic improvement with 20000 pts and 100 iterations.

Could someone explain why ?

Thank you very much,

Best regards

Maia

[Marcel Luethi]

Dear Maia

It seems weird that there are no points on the nose. Are the points that you are showing the ones from the sampler, or are these the points that you obtain after finding the closest point?

If it is the latter, the reason might be that the the points on the nose are just never the closest points, because the nose in the model you start with is far away from your target. There are several options to mitigate this:

1) You can click a landmark on the nose and repeat the process with the posterior model

2) You could extend your search when searching for the closest points (e.g. you choose the point that is closest in normal direction)

3) You start using MCMC Sampling (Tutorial 14 and 15) instead of using pure ICP

Option 3 is what we are using in our applications. It is a much more powerful approach and alleviates most problems that a pure ICP algorithm has.

Regarding the accuracy when you increase the number of iterations and the number of points: You might just always land in a local minimum, irrespective of how many points you take. Thus increasing the number of iterations and points will just bring you more and more closely to that local optimum, but never to the global one.

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/ae0b0209-907c-48a2-86c0-dccdd26121d7n%40googlegroups.com.

[V R]

Hi  Marcel,

You are  right. I displayed the corresponding points. I have points on the nose initially. 

I will follow your advice with MCMC approach.
Thank you very much.
Best regards

Maia