Advanced Setups 11 – DIY Laplace & Poisson Solvers: Jacobi, Gauss-Seidel & SOR
🕑 Added 2021-11-25 08:08:29 +0000 UTCLet’s start tackling the bigger math (and physics and compsci) problems! In this video we’re gonna start building towards a decently fast Poisson solver which we can later use to not only generate minimal surfaces, turn normals into a heightmap or calculate how heat flows through material, but also to design caustics.
This one is based on Matt Ferraro’s genius article on Poisson’s equation and how to solve it numerically.
Comments
Konstantin Magnus
Hi Sam, no I haven't. Would love to redo this process at some point, though!
Sam Beebe
Did you do this in Houdini? Would love to see, as there isn't a built in or easy way of doing this currently!
Nelson Noa
Very nice!
Entagma
Interesting - makes sense! Thanks for elaborating! :) Cheers, Mo
Konstantin Magnus
Hi Mo, the devs explained it like this to me: you enclose a 3d scanned point cloud with a few very large voxels that get iteratively subdivided for as long as they still contain points. That way you will end up with a watertight voxel field, that can be turned into a surface mesh using the marching cubes algorithm. Poisson was basically used before the meshing to fill and blend any kind of holes (meaning voxels that have remained large due to lack of detail in the point cloud), mostly from scan shadows.
Entagma
I'm intrigued - how did poisson show up in that surface reconstruction technique? Cheers, Mo
Konstantin Magnus
Really curious to see where you are going to take this! I stumbled upon Poisson first time at work for surface reconstruction, specifically voxelizing point clouds along with marching cubes to create watertight meshes.
Cirno
love this topic!!!