3D poisson surfacing of pointclouds

[freaq]

has anyone attempted to do this in Houdini or other 3D surfacing algorithms?
can anyone speak to the feasability of this? I'm thinking of attempting it, but could use some pointers as this is quite complicated and I have very little experience (basically none) with surfacing...

 

Greets freek

[Wooshum]

I became interested in this but never attempted an implementation within houdini. I did bookmark this article which you may find helpful:

 

http://devmag.org.za/2009/05/03/poisson-disk-sampling/

[ikarus]

I did a vop solution of poisson surfacing a while back but it had some problems with more complicated geometry and wasn't any faster than the available sesi solutions for surfacing. 

[freaq]

the main reasons I want it is to actually use the points I have specified I can;t simply do a limit surface or anything due to topological constraints...

[Serg]

You can get something that looks like poisson by just scattering very dense point cloud then fusing points within a radius (with fuse sop, set to keep unused poitns).

 

sort of OT: I really wish hair/fur used poisson for its scattering, because we normally have increase fur density a lot to compensate for the gaps/holes visible in the scalp caused by the regular random distribution... 

[freaq]

well the biggest problem here is that I have no surface and I need to create one from fairly arbitrary topology..

[edward]

Have you tried the Point Cloud Iso SOP? If you have no normals, try maybe petz's file from this thread: http://forums.odforce.net/topic/13655-point-cloud-normals/

 

EDIT: pclaes had an updated .hip for H12 here http://forums.odforce.net/topic/13655-point-cloud-normals/?p=100545

[magneto]

So something like this?

 

 

I wish Houdini had more functionality like meshlab.

[freaq]

Exactly like that!!!

[anim]

if it's exactly that (more or less planar projection surface)

then you can just save rest position

do Triangulate2D

and copy rest back to P to get the same result

 

for objects with cavities though, it's much more difficult

[freaq]

well the example actually shows some overhangs in that model,

the delauney triangulation 2D triangulate would not be able to handle that pointcloud I'm afraid.
I have considered, "smoothing my pointcloud" to something that could be mapped as a 2D manifold when storing positions, then Triangulate and then replace,
but overlaps and tehrefore non manifold geometry are just not solvable that way. I could do almost like a UV unwrap, but it would be painful.

Edited July 24, 2014 by freaq

[freaq]

I mean look at this for example:

this would be incredibly usefull in Houdini
I wonder whether it would be possible as Meshtools are opensoucre to potentially integrate this into a sop... 
I lack the knowledge to do so unfortunately.

[acey195]

Have you tried converting vdbfromparticles?

and possibly a convert VDB (back to polygons) afterwards?

 

something like the example.

 

vdbExample.hipnc

Edited July 27, 2014 by acey195

[edward]

I'm surprised that no one has yet mentioned to just run the command line meshlab from a Unix SOP since it can read/write .ply files. So you can easily just have a command that converts to ply, run through meshlab, convert back to bgeo.

[Jason]

I've wondered about the feasibility of writing plugins using PCL (from http://pointclouds.org) for Houdini -- Limited python bindings: http://strawlab.github.io/python-pcl/

 

I've never done it though

[freaq]

I'm surprised that no one has yet mentioned to just run the command line meshlab from a Unix SOP since it can read/write .ply files. So you can easily just have a command that converts to ply, run through meshlab, convert back to bgeo.

 see that would be perfect, I was thinking of going through the sourcecode or something... or creating my own, but just seemed so hard and complex...

I might give that a try thank you!

and Jason if you ever do give that a try let me know please!

the problem with VDB is that it does not use the actual points but rather envelops,

one could do some smart tricks there, but basically you always have 2 surfaces instead of one.

and the surface s away from the border, and is susceptible to holes...

Edited July 29, 2014 by freaq

[edward]

Personally, I thought my first post in this thread was sufficient. <shrugs>

[freaq]

Personally, I thought my first post in this thread was sufficient. <shrugs>

afraid the point cloud ISO does not do what I need :/

but the unix meshlab commandline sop trick might...