Target.ply in datasets

64 views
Skip to first unread message

Vénérée Randrianarisoa

unread,
Oct 6, 2020, 6:16:37 PM10/6/20
to scalismo
Hi,
I'd like to know if  the mesh target.ply is has the real nose of Dr. Thomas Vetter.
Thank you
Maia

Marcel Luethi

unread,
Oct 8, 2020, 2:40:17 AM10/8/20
to Vénérée Randrianarisoa, scalismo
Hi Maia

Yes, that is his real nose. But feel free to reconstruct a new one for him :-)

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/ea159ca7-d650-43dd-8be8-69bca1d441fdn%40googlegroups.com.

V R

unread,
Oct 8, 2020, 6:10:37 AM10/8/20
to Marcel Luethi, Vénérée Randrianarisoa, scalismo
Thank you very much.
The one I reconstructed is almost the same. Awesome !

Maia R.

unread,
Oct 13, 2020, 12:03:37 PM10/13/20
to scalismo
Hello Marcel and dear community, 
I'd like to compare the nose model I get wrt to the original nose (target.ply). They are not registered.
I see that in:
can be used to compute such a distance.
As far as I understand, the two meshes must be registered. Do they need to be in correspondence (i.e. point by point) before computing those metrics or a rigid alignment with a few orresponding landmarks are enough ?
Also, to clip/marginalise the nose, I thiuygh to use this sample codes as in Shape completion tutorial (Tutorial 9):
=============
 val nosePtIDs = model.referenceMesh.pointSet.pointIds.filter { id => (model.referenceMesh.pointSet.point(id) - model.referenceMesh.pointSet.point(PointId(8152))).norm <= 42 }
=============

but instead of Point IDs (which assumes that the original nose and the nose model are fully in correspondence), I would use point coordinates.

Does this sound a goo approach ?

Thank you very much

Marcel Luethi

unread,
Oct 14, 2020, 3:32:55 AM10/14/20
to Maia R., scalismo
Dear Maia,

The meshes don't have to be registered to compute the metric. Most mesh metrics simply work by taking the closest point on the surface as the corresponding point. You may, however, want to clip the reference such that it only shows the region you are interested in before computing the distances. You can do this in a tool like meshlab or paraview.

Best regards,
Marcel


V R

unread,
Oct 14, 2020, 7:02:23 AM10/14/20
to Marcel Luethi, scalismo
Hi Marcel,
Thank you very much! I got impressive results !
Best regard
Maia

V R

unread,
Oct 14, 2020, 8:53:41 AM10/14/20
to Marcel Luethi, scalismo
Hi Marcel, 
I'm not sure to understand. While you're saying not registered, do you mean point to point correspondence ?

With rigid alignment by means of a few landmarks, I got much better average distance compared to not aligned meshes.
avgDistRegistered  = 1.3701001748428243
diceRegistered = 0.9801405668258667

avgDistNotRegistered  = 205.12343229549083
diceNotRegistered = 0.0
The diceNotRegistered is equal to 0 since the two meshes have no common parts, that's why I am not sure to fully understand.

image.png
Best regards

Marcel Luethi

unread,
Oct 14, 2020, 10:54:38 AM10/14/20
to V R, scalismo
Hi Maia

This was my mistake. I apparently did not read your email properly and thought that you mean that the meshes need to be in correspondences. While you don't need correspondence, the meshes should be aligned as you suggest.

Best regards,
Marcel

V R

unread,
Oct 14, 2020, 5:29:42 PM10/14/20
to Marcel Luethi, scalismo
No problem. Thank you anyway.
Maia
Reply all
Reply to author
Forward
0 new messages