How to avoid double walls created by convertVDB

[konstantin magnus]

I have a point cloud with large open parts and turn this into a VDB from particles.

Logically, but unfortunately I get double walls after meshing it.

Whats the most robust way to get a hollow model from this right away? Even if there are dozens of parts, messy geometry etc.

PC_double_wall.hiplc

Edited April 19, 2018 by konstantin magnus

[Noobini]

here's my go...but it needs the Normals to be established for Peak to work tho...if you were given just point cloud with no normals...dunno...

PC_double_wall_vu.hipnc

(but double walls are good for massage parlours tho)

Edited April 19, 2018 by Noobini

[konstantin magnus]

Hi vusta, thanks for your help.

The point cloud`s normals are potentially corrupted, too. So, as you suspected, I am looking for a more technical approach that may include SDFs gradients or possibly any information coming from within the VDB field.

[MENOZ]

i tried this road but i'm stuck:
i am trying to find the borders of the mesh with an attribute (I tried with the curvature and with another silly method, both gives the same results in this case)  and primitiveSplit them so in my mind i would have the big face on one side, the faces at the borders, and the other side of the mesh. The problem is that this method is very dirty and i would need to create a clean seam line. Once you can separate the 2 big faces you can choose the one you prefer based on some parameters (like occlusion maybe?) and reproject it more or less in the position where the original scattered points were. 

The problem is to find a cleaner border edge to use to split the prims. 

maybe there is a good way to find the borders of the original scattered points, I usually count how many neighbours in a radius and use a treshold to filter those particles.. but is not always great, and use those border points to create a line, then extrude and boolean the mesh to split it.

I attach the file if you want to have a look... :/  I'm pretty sure there must be some mathy way of finding those features.

findBorders.hipnc

[Atom]

Would a delete by color work at the end?

 

ap_PC_double_wall.hiplc

[konstantin magnus]

1 hour ago, MENOZ said:

occlusion maybe?

Yes sure, I can delete points by their occlusion. But my question is rather if creating inner walls can be avoided right at the outset?

Occlusion_test.hiplc (make sure to enable game dev tools)