CloudCompare forum

Hi,
I am relatively new to CC and I am trying to assess the quality of some meshes I generated.
I have used C2M to calculate signed distances, however they don't provide information on missing parts in my generated meshes.
What I mean by that is: If I would have a generated mesh that is only half of the reference mesh (for example if my reference was a cube and my generated mesh only half the cube) but that half would fit perfectly on the reference mesh the mean and standard deviation would be perfect but the generated mesh would not be a good representation of the reference mesh. Hope this makes sense. How can I include this in my similarity assesment? My first impulse was to flip the compared and reference mesh and somehow compare the two results but I wouldn't know how to compare them and I read in another question that the process is not symmetrical.
If anyone has ideas on how to account for missing pieces or holes in my generated mesh I would be very thankful!

Well, your initial hunch doesn't seem that bad ;). Comparing both ways would at least let you detect the issue.

Otherwise, maybe the 'Closest Point Set' could help you as it returns all the points that are the closest to another point cloud. So if the number of 'closest points' is much smaller than the total number of points, then you could also detect an issue. The issue is that it only works with point cloud (so either you can use the mesh vertices directly, or you may have to sample points on your meshes).

Thank you for your reply!
I think I will do the "compare swap" and then take the mean distance of the swapped comparison as an indication for holes or missing parts. In your opinion, if I want to choose the "best fitting mesh" to my reference, is the mean distance or the standard deviation more important? I want to take them both into concideration as well as the "hole index" and weigh them differently according to their importance.

Well, it depends on how the 2 meshes are aligned:
- if they are aligned in an absolute coordinate system, then the average would give you a real information (such as an absolute translation, or a change of volume/scale).
- but if they are best-fitted/aligned with each other, then the average is meaningless (it should ideally be close to zero, unless you have big discrepancies between the 2, but the std. dev. will already give you that information in a much better way

I used the align points method for all my meshes (ICP gave me kind of random results) and never changed the position/scale of the reference. Doesn't that mean that they should have the same coordinate system?