Repulsive Shells - Conference Presentation
Vložit
- čas přidán 10. 09. 2024
- This video gives a short overview of the SIGGRAPH 2024 paper "Repulsive Shells" by Josua Sassen, Henrik Schumacher, Martin Rumpf, and Keenan Crane.
For more information, see:
www.cs.cmu.edu...
Abstract: This paper develops a shape space framework for collision-aware geometric modeling, where basic geometric operations automatically avoid interpenetration. Shape spaces are a powerful tool for surface modeling, shape analysis, nonrigid motion planning, and animation, but past formulations permit nonphysical intersections. Our framework augments an existing shape space using a repulsive energy such that collision avoidance becomes a first-class property, encoded in the Riemannian metric itself. In turn, tasks like intersection-free shape interpolation or motion extrapolation amount to simply computing geodesic paths via standard numerical algorithms. To make optimization practical, we develop an adaptive collision penalty that prevents mesh self-intersection, and converges to a meaningful limit energy under refinement. The final algorithms apply to any category of shape, and do not require a dataset of examples, training, rigging, nor any other prior information. For instance, to interpolate between two shapes we need only a single pair of meshes with the same connectivity. We evaluate our method on a variety of challenging examples from modeling and animation.
amazing presentation and the camel killed me
I literally went, "OOH!" when I saw that glove turn inside out. Incredible tech.
Please never stop doing research of such high quality.
Truly revolutionary! I'm a huge fan, I really appreciate your passion and commitment to educating; it inspired me as I finished my doctorate and I adapted your presentation style for own defense. Your work has led me down an incredible rabbit hole that I doubt I will ever exit.
So glad to hear that. Godspeed.
I'm not an expert but this is one of the most amazing thing that I saw.
It feels like revolution.
I want to understand clearly its math.
I could follow this just enough to remain very impressed, but I'm leaving with plenty of inspiration for what I'd like to study further. Putting this on my "watch later" playlist and hoping next time I'll feel ready to read the paper :)
Maybe the best presentation I've ever seen!
I am in awe. thank you for the amazing presentation and visuals as always. huge shoutout to the coauthors
Awesome! I wish all lectures were as fast and insightful as this; but obviously it takes exponentially more time to prepare.
This might be the coolest shit I've ever seen in my life
Nice! Path-of-least-resistance fans flow in, this is hot.
dam. i love the idea of my video game models moving more lifelike when grabbing things.
also has cool applications for robots picking up household items
Beautiful presentation
oh man, that plan view of the leaf (around 9:30) looks so similar to an Euler spiral
Brilliant! Inspiring as always. 👏
Repulsive curve →Repulsive surface → Repulsive shells. nice work!
Great presentation. Would be nice to see real world applications for 3D artists, though your examples speak a lot for themselves already.
I guess the one application that was implicitly mentionned is for interpolating pose to pose animation in a non colliding way which would be a huge feat and would save a lot of time.
Very interesting! I feel like this has a lot of potential for practical application. The multipole-guided adapative refinement all by itself seems like a useful building block for new approaches to old problems.
It would be fantastic if you could share whatever code you made. If you have it, I personally would love Houdini files, but of course, anything would be appreciated. (At least I didn't see anything linked.)
@@DimiShimi We are indeed putting together a code release. No plans for Houdini directly, but I know it’s fairly straightforward to wrap external libraries into Houdini.
@@keenancrane Do us all a favour and make it C.
Amazing presentation, Prof. Crane!
Goat professor
That laundry demo was excellent!
Great presentation. The applications you've shown here are truly amazing! Especially loved the packed octopus. I hope to see more of this technique in the future, will be following closely!
Amazingly delivered! Deforming in anticipation of obstacles reminds me a lot of Quantum Physics. I wonder if perturbation theory can be used to adapt the graph manifold to medium-range environmental changes (moving obstacles, for example)
Great work. Smallest sphere curvature makes sense. I would love to see a longer, more detailed demonstration of the whole mapping process.
Amazing!
Really interesting solution! This can be great for soft robotics
Add gravity and balance, and you might finally be able to make natural walking humanoids! Walking, running, jumping,
Simulate explosive motions, muscle fatigues, in the future robots will do parkour! ❤
I wonder how this might work for mapping partial deformation of mesh interactions. Blacksmithing comes to mind.
You are a genius!
This shape space would be somewhat equivalent to the moduli space of meshes and such?
Something I don't discuss here is that the configuration manifold M can be formally viewed as the quotient of surface immersions modulo rigid motions (since the energies are rigid motion invariant).
9:05 is the craziest thing I've ever seen. Insane. How big can these meshes be? Could this be distributed across thousands of cores?
It is fascinating to see how differential geometry is applied in computer graphics, and it would be awesome to analyze how these ideas get implemented in discrete settings. Can someone recommend an open-source project which used these ideas?
I teach a course on connecting differential geometry to discrete algorithms, including applications in computer graphics and geometry processing: geometry.cs.cmu.edu/ddg
A lot of these ideas are implemented in the open source software Geometry Central: geometry-central.net/
Woah!
this is so cool. are you familiar with the shape spaces and infinite dimensional manifolds by michor and his collaborators? It looks somewhat similar
@@Czeckie Yes indeed. Our work builds on this topic, and especially work on (infinite dimensional) shape spaces of surface immersions.
keenan tha goat
I wonder if, with some modification, this could be used to find the embedding of n-manifolds in 2nD space... I wonder if there's a way to prune discrete n-manifolds such that their topology is unchanged, but converges to a minimal representation...
What courses to take in order to understand all this ? I have elementary calculus and linear algebra
There's no one perfect course, but this one may be a good starting point if you've already done calculus & linear algebra: geometry.cs.cmu.edu/ddg
for the male body mesh, do the two poses have to have the same topology? If not, then could this be used for re-top purposes?
@@jasonford7439 Yes, we assume fixed mesh connectivity (though that’s a restriction we’re very interested in lifting…)
@@keenancrane I can't find it anymore but years ago I saw a video/paper that used poisson techniques to retopologize a set of head scans jointly such that each head ended up with the same topology, allowing them to blend them in various ways. I think they did the same with a rabbit and horse mesh, blending those. Not sure if that's enough for you to go on.
@@jasonford7439Yep, anything that gives consistent topology should work as input. A bigger challenge is that if you want to generate trajectories that experience significant deformation somewhere in the middle, we need to be able to dynamically adapt the topology (so that numerical quality doesn’t suffer).
really cool wow
word
this is so cool, any chance the code will be available for us who can't follow the math?
@@ViperPacket It will happen…
@@keenancrane What math classes would be good to take to be able to understand this??
@@picosdrivethru Good question. Probably: differential geometry, and a little bit of calculus of variations (though you may get what you need already in a differential geometry class). Also a bit of solid mechanics/elasticity, and numerical analysis. But it’s not like you need to take full blown classes in all four of these subjects. (I didn’t!)
Siggraph is always wizard shit
hmm, but this approach merely determines transformations where contact is avoided? it still doesn't look real. the hands are a good example - no one would be able to clasp their hands like this without touching. you would need to find some sort of balance between the transformations determined by this algorithm (which look great as a starting point) and some sort of deformation on contact, I think? like maybe you could have "repulsive bones" inside a hand model, while still allowing "skin" surface contact with deformations?
This sounds an awful lot like UMAP
except it's not remotely uniform or linear