The following is the particle cloud on frame 1 loaded in a PRT Loader. It contains only 12,302 particles. I had to inject a Header into the CSV file since it contains Position X, Y Z and Mono Color (which I imported as “Density”):
This is the FROST mesh without the color color display. I used Radius of 2.0, Zhu/Bridson with Blend Radius 3.0, default Trimming, default Meshing res of 1.5:
I added an Omni light to the scene, enabled Hardware shading with all effects (Soft Shadows, AO) and could get interactive lighting (looks much better in motion):
Ive got a Kinect up and running and meshing.
Been using it to do crude pointcloud set surveys for complex tracking.
Frost indeed works well with the data - a recent update to the Brekel software im using means faster writing to disk of pointdata, resutling in performanace capture of sorts, which looks pretty cool with frost - talking liquid faces
il post some results when i have time. i did post an overview of the Kinect and resutling meshes on my vimeo page
Krakatoa ships with a PRT Source object which can be used to read a PRT stream from a stand-in scene object (a 3rd party object implementing the same data stream all Krakatoa objects and Frost use internally).
So I think it would be better to write a Kinect stand-in object for Max that would get the particles from the Kinect driver and format them in the PRT stream format in memory, thus avoiding the writing to intermediate files.
The necessary code to implement such a stream is available. I think some of the Thinkbox developers (Paul, Darcy) could help you with that…
Thinking about this in the shower this morning… Wouldn’t the isosurface be offset from the LIDAR data by radius units?
Sometimes particles represent the center of something. Like for SPH or metaball modelling or whatnot.
But in the case of LIDAR or other range scanning systems, the points are AT the surface. Same thing as if you were to mesh the vertices of a mesh. As it is now, you’ve made Thom Yorke’s nose too puffy, right?
What’s the proper way to mesh something like that? Obviously doing something where building some sort of connectivity network by finding say the 2 closest points and making a triangle would work, but using the Frost-style marching cubes what’s the closest we can get? Perhaps moving the final mesh vertices -Normal*weightedRadius or something?
Even in the case where you want to turn 1,200 particles into 12,000,000, you’d have this same issue… You want the resulting particles to be built inside the bounds of the original particles. Or maybe not?
I suppose the answer may be to just store the weighted radius into the mesh and we can handle it in the modifier stack. Haven’t tried any of this, didn’t have Frost in the shower. I might not be making sense.
You are totally right - Tom Yorke is quite skinny, Frost added a lot of “fat” to his face (mainly because the LIDAR resolution in that case was very low).
When meshing that GMC model though, the distance between the particles was so small that the particle radius was almost negligible - retopologizing the blob surface would have produced a model VERY close to the original surface, and a little push of the retopo mesh along the normals by the radius amount would probably produce something matching the source.
But keep in mind we discovered this application field of Frost just weeks before release, so there is a lot that could be done for future releases
Btw, I discovered yet another application area of Frost this morning and will present it at NAB, then probably post a tutorial about it. I was totally surprised
I suspect there are other things we haven’t thought about …
I’d still like to see density weighting as well as negative radii in Frost. Combined with signed distance from the PRT Volume (please?), we should be able to do fairly accurate surface-surface reconstruction. Maybe not ideal compared to working in voxels directly, but we gain Krakatoa toolset in between, which could be a big plus. A lot of our data comes from placement of points at the surface threshold, which works great for Krakatoa, so we’d really be interested in how we could maintain good surfacing there.
We’re getting lovely use out of Frost as a “PRT Boolean” tool as well as a means of generating normals from point data that for whatever reason didn’t originally have it. Neither of those uses were primary motivators for us, but it’s a great workflow enhancement.
I can’t believe you are going to make me go to Vegas to see it! There is a bunch of stuff going on at NAB this year (things that normally aren’t there) I dropped going to NAB for Siggraph since it is in Vancouver, makes me wish I hadn’t.
