Alexandre Binninger | The Graphics Frog

Thank you! Really appreciate it 😄

TetWeave: Isosurface Extraction using On-The-Fly Delaunay Tetrahedral Grids for Gradient-Based Mesh Optimization TetWeave is a novel isosurface representation that jointly optimizes a tetrahedral grid and directional distances for gradient-based mesh processing like multi-view 3D reconstruction. It dynamically b...

🔥 Want to try it?

🛜 Project page: alexandrebinninger.com/TetWeave/
💻 Code: github.com/AlexandreBin...
📄 Full paper: arxiv.org/abs/2505.04590
(9/9)

🛠️ TetWeave can be used in many applications:
✅ Mesh compression
✅ Geometric texture generation
✅ Photogrammetry
(8/9)

🏆 Results? TetWeave guarantees watertight, manifold, and intersection-free meshes, and outperforms FlexiCubes and DMTet in memory efficiency and mesh quality. (7/9)

👁️ Because we compute the grid on the fly, the point cloud can be located anywhere! We develop a resampling strategy that allocates more points where they matter most (like highly detailed regions) based on rendering error or customizable approaches. (6/9)

🎛️ We introduce a notion of directional signed distance, encoded as spherical harmonics on each grid vertex. This allows precise vertex placement and sharper details! (5/9)

💡 TetWeave solves this by dynamically constructing the grid via Delaunay triangulation! Our representation simply consists of a point cloud with an SDF value. To reconstruct a target mesh from images, we jointly optimize the grid and the signed distance at each point.
(4/9)

🔍 Methods like DMTet or FlexiCubes jointly optimize the implicit representation and the grid structure from which the final mesh is extracted. However, they rely on grids with fixed topology, which leads to memory inefficiency and poor scaling. (3/9)

🤔 Meshes are ubiquitous in computer graphics. However, their fixed topology and the difficulty of recovering from a bad state make them a poor choice for gradient-based optimization pipelines, such as 3D reconstruction. (2/9)

🔷🦀 TetWeave: Isosurface Extraction using On-The-Fly Delaunay Tetrahedral Grids for Gradient-Based Mesh Optimization is accepted to #SIGGRAPH2025 (TOG).

Done in collaboration with @rubenwiersma.nl, Philipp Herholz, and Olga Sorkine-Hornung.

But why a new shape representation? (1/9)

Is this art of AI failure?🤔

I remember playing this game! Such a great one!!

That's so cool and creative!

For me it’s somewhat in the middle of afternoon, so I tend to cherish this time slot to go to gym :D

PorcelAIn!

Will AI be able to create porcelAIn pieces of artwork? 😉 At least it'd make a good AI company name

Inspired by Clairen from Rivals of Aether II, I generated this futuristic black cat with SD-πXL.

It turned out to look a bit evil, so I tried with a dog and... yeah, it feels different.

Which one would you choose to side with?

What did this cat see on its laptop to look so shocked?

Really beautiful! Is it done by hand? 😯

GitHub - AlexandreBinninger/SD-piXL: Repository for SD-πXL: Generating Low-Resolution Quantized Imagery via Score Distillation (SIGGRAPH Asia 2024) Repository for SD-πXL: Generating Low-Resolution Quantized Imagery via Score Distillation (SIGGRAPH Asia 2024) - AlexandreBinninger/SD-piXL

SD-πXL is available here: github.com/AlexandreBin...

A few images made with my latest project, SD-πXL. All with a limited amount of colors!