Building on our SIG Asia "Reach for the Spheres" idea, we interpret discrete signed distance field data as a collection of spheres, to which the unknown surface must lie tangent at (at least) one point. We further observe that only the "arcs" of spheres not covered by other spheres can host valid closest/tangent points.
Check out my latest collaboration with Tetsuya Takahashi on efficiently solving box-constrained QPs, especially for fluid animation problems on Cartesian grids, using an interior point method and a tailored algebraic multigrid preconditioner: https://tetsuya-takahashi.github.io/PD_BCQP/
And here's the YouTube link: https://www.youtube.com/watch?v=GeeweiNzcOQ
So far this year a quick Google turns up only two SIGGRAPH 2024 authors that have prematurely broken anonymity on their websites before receiving final acceptance of their paper. That's not half bad given the number of papers accepted.
Our paper on solving Poisson problems using Monte Carlo raytracing-like techniques now has a video up on youtube, thanks to first author Ryusuke Sugimoto: https://www.youtube.com/watch?v=5SDhXJ3Sjmo
Actually pretty excited to attend my first ever SIGGRAPH Asia next week! Only took me 15 years to get around to it.
I'll be chairing the Fluid Simulation session on Tuesday and @silviasellan is presenting our paper on SDF reconstruction at the Reconstruction session on Thursday. See ya there.
I like when academics say "I'm looking for a postdoc" because you don't know if they want to be one, want to hire one, or just misplaced one somewhere.
For computer graphics academics and researchers, I threw this list of folks together. Apologies if I've forgotten anyone or included anyone who prefers to be forgotten! Let me know if there are any errors or corrections. Hoping this will help folks migrating from the birdsite. 🙂 https://christopherbatty.github.io/Mastodon-Graphics-Academics/
I'm very excited to share our new SIGGRAPH Asia 2023 paper on improved reconstruction of explicit surface meshes from signed distance field data: "Reach for the Spheres: Tangency-Aware Surface Reconstruction of SDFs".
The key idea of our "Reach for the Spheres" algorithm is to interpret each SDF sample point and its associated distance value as representing a sphere (or circle in 2D). Then, a valid reconstructed surface that satisfies the SDF data must never enter any sphere, the sign of each sphere dictates which side of the surface it lies on, and, critically, the surface must touch each sphere (at least) once.
For example, compare here a Marching Squares reconstruction vs. a valid surface.
By seeking to satisfy these conditions, our algorithm can provide superior reconstructions (e.g., vs. Marching Cubes, Neural Dual Contouring) on incredibly low resolution SDF grids, such as this 10^3 grid example.
This was a fun collaboration with Silvia Sellán (Toronto) and Oded Stein (USC).
By the way, first author, U of T PhD student Silvia Sellán is on the job market this year, so be sure to check out her three other brand new papers, and also hire her!
For those attending #SIGGRAPH, be sure to get up early for the "Pushing the Boundaries" paper session on Wed at 9am. Ryusuke Sugimoto will present our work on a Monte Carlo raytracing-like algorithm for solving Laplace/Poisson problems called the Walk on Boundary method. https://rsugimoto.net/WoBforBVPsProject/ Then, you can start converting all your favorite rendering techniques into PDE solvers!