Sean (Aug 13 2018 at 18:03): Sean (Aug 13 2018 at 18:03):@starseeker Just listened to an outstanding paper you'd love. the work fully generalizes and improves upon SPSR, amazing results for healing/sealing meshes AND point clouds.
starseeker (Aug 14 2018 at 03:17):@Sean Very interesting indeed! At a casual glance I'm not immediately seeing how they got from the winding number to the output mesh - the github code seems to not include that part, and the paper mentions something about "a continuation method [Wyvill et al. 1986] for voxelization and isosurface extraction." That doesn't ring any bells for me - is it something you're familiar with?
starseeker (Aug 14 2018 at 03:20):If it's something along these lines he has a C implementation in the back of the paper... http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.24.3429&rep=rep1&type=pdf
starseeker (Aug 14 2018 at 03:24):https://github.com/Tonsty/polygonizer
starseeker (Aug 14 2018 at 03:29):I may have to kick the tires on this and see if I can get it to do something...
Sean (Aug 15 2018 at 17:34):@starseeker I'm not sure about continuation methods, but Wyvill is a familiar name, did a lot of good CSG papers in the 90's like http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.409.5011&rep=rep1&type=pdf
I suspect they're referring to the continuation method in http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.881.7066&rep=rep1&type=pdf
Sean (Aug 15 2018 at 17:37):If it's something along these lines he has a C implementation in the back of the paper... http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.24.3429&rep=rep1&type=pdf
ah yeah, so you found a follow-on implementation. that looks like the right one.
the interesting part about the whole winding number paper is that it's essentially controlled by a single parameter, something akin to a smallest feature/hole size, similar to ball pivoting but with even better sealing/healing properties.
Sean (Aug 15 2018 at 17:38):the presentation was pretty great too, might see if we can get ahold of the slides
Sean (Aug 15 2018 at 23:17):also worth noting if you didn't notice, there are separate repos for point clouds and triangles: http://www.dgp.toronto.edu/projects/fast-winding-numbers/
starseeker (Aug 16 2018 at 01:28):Saw that - we'll need to adapt the triangle code, since the point version uses some CGAL routines. I was looking at that yesterday - I think I have a rough sketch of what needs doing, but it's some work to tie the pieces together.
starseeker (Aug 16 2018 at 12:33):@Sean If you get a chance to ask (or if it was covered in the talk) how does this approach do with sharp corners?
Sean (Aug 25 2020 at 16:33):@starseeker Listening to a really interesting "Advanced CAD Data Preparation" session that is essentially a graphical drop-down form-based search+exec recipes that are applied to geometry.
starseeker (Aug 25 2020 at 16:33):cool!
Sean (Aug 25 2020 at 16:35):Only thing it has that I don't think we have yet is a way to filter down based on geoemetric properties and sub-properties like the bounding volume or number of mesh verticies if it's a bot, etc .
Sean (Aug 25 2020 at 16:39):It's really encouraging hearing a technical salesman go into detail about how hard, advanced, awesome that software is that it can now do that, and it's something we've had for years. Granted, they have a pretty point-clicky GUI.
Sean (Aug 25 2020 at 16:41):haha, and now he's talking about shooting rays at vertices to determine inside/outside visibility (in order to cull internal objects)
Sean (Aug 26 2020 at 19:45):https://blog.selfshadow.com/publications/s2020-shading-course/
Notification Bot (Aug 06 2023 at 17:01):Sean has marked this topic as resolved.
Notification Bot (Aug 06 2023 at 17:02):Sean has marked this topic as unresolved.
Sean (Aug 06 2023 at 17:03):Collapsed the two siggraph streams (SIGGRAPH 2020 and siggraph goodies) into one SIGGRAPH stream.
starseeker (Aug 06 2023 at 17:04):Ah yes, it's that time again :-)
Sean (Aug 06 2023 at 17:04):Really awesome course on Neural Fields, all materials available online. I'd read about them last year when they were unveiled, but really hits home lots of use-cases...
Sean (Aug 06 2023 at 17:06):Could see it having some use in geometry conversion, going to/from representation formats (e.g., CSG to BoT or point cloud to NURBS to CSG, etc).
starseeker (Aug 06 2023 at 17:07):What's the necessary technical infrastructure like? Big dependency stacks?
Sean (Aug 06 2023 at 17:09):depends on whatever neural network framework is used, huge variety from big guns like tensorflow to really streamlined projects like dlib or completely home-grown for a specific network (e.g., yolo)
Sean (Aug 06 2023 at 17:30):Might be a really easy way to optimize ray tracing performance for expensive geometries. We could bake a model into a neural net representation (to any degree of fidelity), then use that during ray tracing.
Sean (Aug 10 2023 at 17:42):so close! ... excellent paper on real-time pixel-accurate NURBS tessellation, but only for viz
Sean (Jul 24 2024 at 18:25):Priming this channel to the top of the list for interesting/relevant SIGGRAPH 2024 research.
Sean (Jul 26 2024 at 15:51):@starseeker Jeff said to be sure to tell you he says hello, best wishes, fond memories. Plan to deep dive with him a bit on how we might be able to leverage Anari (or OSPRay under Anari (or Appleseed under OSPRay under Anari!))
starseeker (Jul 26 2024 at 16:42):Same here - please give him my regards as well!
Sean (Jul 27 2024 at 22:57):Nice kudos and recognition from Peter Shirley at HPG24 to BRL-CAD. He gave a keynote talk, a great anthology on the history of real-time rendering and computer graphics.
Sean (Jul 29 2024 at 17:11):HPG Day 2
7:09 Jeff Amstutz talk on ANARI
7:52 Peter Shirley's keynote
Shirley's entire keynote is an outstanding history of CG and worth watching, but his BRL-CAD mention is around 8:06:30.
starseeker (Jul 29 2024 at 19:44):Good talks. Apparently as a BRL-CAD guy I qualify for the "old timer's reunion" :-P
starseeker (Jul 29 2024 at 19:49):Correct me if I'm wrong, but ANARI sounds like it would be something to put on top of our libraries as a layer between our rendering and MGED/QGED/Archer/Arbalest?
starseeker (Jul 29 2024 at 19:50):Doesn't sound (at least from that brief talk) like it would standardize a way for us to insert our ray intersection math solving into another engine.
starseeker (Jul 29 2024 at 19:53):It does sound potentially appealing to me though as a way to organize requesting rendered images from BRL-CAD's libs
Sean (Jul 29 2024 at 21:08):That's right. We'd either put librt under anari as a renderer, or librt under appleseed under anari, or librt as an ospray plugin (which is already under anari). The benefit or idea being we'd be able to be a renderer option in any anari-adopting application without having to do the glue ourselves N times over for every app+renderer context.
Sean (Jul 29 2024 at 21:10):One of the most interesting and relevant papers thus far... https://arxiv.org/pdf/2404.13420
There's video and code -- and it appears to newest state-of-the-art beating out SPSR and others pretty heartily and going the xtra step to produce CAD geometry.
Sean (Jul 29 2024 at 21:11):https://github.com/QiujieDong/NeurCADRecon
starseeker (Jul 29 2024 at 21:11):python, CUDA and pytorch as requirements? Oof
Sean (Jul 29 2024 at 21:14):starseeker said:
Good talks. Apparently as a BRL-CAD guy I qualify for the "old timer's reunion" :-P
There were a few hands that shot up in the room. :)
20+ years gets one the Pioneer status from ACM -- you're not too far from that too!
Sean (Jul 29 2024 at 21:16):starseeker said:
Correct me if I'm wrong, but ANARI sounds like it would be something to put on top of our libraries as a layer between our rendering and MGED/QGED/Archer/Arbalest?
Jeff actually recommended we join the advisory panel, to ensure we get the hooks needed (e.g., to directly render nurbs or CSG) which could also mean adding hooks so we could inject ourselves as an intersection engine under other renderers like we did with appleseed. Then we could render brl-cad geometry automatically in any anari-certified renderer.
Sean (Jul 30 2024 at 21:07):Another excellent paper -- direct conversion of CSG to poly with adaptive feature preservation, vastly improved over the Delaunay tet method. Relatively simple algorithm to implement (alas, we can't use their code, but could use the method): https://jurwen.github.io/Adaptive-grid-for-implicit-complexes/
Sean (Jul 30 2024 at 21:08):Basically, robust conversion for anything that can be raytraced. Will have to read that paper in more detail but results were super impressive.
Sean (Jul 31 2024 at 21:43):Not a siggraph paper, but I found it at siggraph... really cool CAD reconstruction paper involving mesh to CSG. Pretty impressive: https://www.sciencedirect.com/science/article/abs/pii/S0167839624000736
Sean (Aug 01 2024 at 14:06):This is not related to current work so much, but one of the more interesting papers was BrepGen: https://github.com/samxuxiang/BrepGen
Those folks figured out a way for encoding BReps into the latent space so a BRep is directly output from a network.
Sean (Aug 01 2024 at 14:07):Another paper/framework from NVidia making splashes: https://github.com/NVIDIAGameWorks/kaolin
Sean (Aug 01 2024 at 14:07):(encoding as VDB, not BRep)
Sean (Aug 01 2024 at 17:18):This year is insane. Most papers ever (nearly 300??)... and awesome new trend is nearly every paper (4 of 5) has make their source code available. That trend used to be about 1 in 5 previously...
Have notes on over 30 interesting pieces of research we could directly and near-term benefit from.
Sean (Aug 01 2024 at 17:40):Quite a resurgence in CSG research ... so many papers this year
Last updated: Oct 09 2024 at 00:44 UTC