SIGGRAPH 2026 paper maps point clouds to torus fits for fast signed distance queries
keenanisalive · x · 2026-07-29
Nicole Feng’s project page introduces Points as Tori, a SIGGRAPH 2026 paper on fast signed-distance queries for point clouds.
What it does
- Takes a point cloud with normals and produces an analytical parameterization for querying signed distance at arbitrary resolution.
- Uses local torus fitting: a pretrained network predicts per-point curvature and shift parameters, and the method leverages the closed-form signed distance of tori.
- Avoids costly global optimization and spatial discretization, making it parallelizable.
Why it matters
- Unifies signed distance with classic reconstruction ideas like winding numbers and Poisson surface reconstruction.
- Works on point clouds from photogrammetry, meshes, 3D Gaussian splats, and neural implicits.
- Enables direct downstream use without explicit mesh reconstruction, including offsets, Boolean operations, and sphere-traced visualization.
The page also links the paper, posters, GitHub repo, Python API, demos, and training scripts/data.
Related event: SIGGRAPH 2026 Paper PAT Enables Fast Point Cloud Distance Queries(2 posts)→
More from Research
- Analyzing 100k+ Reddit Posts: AI Shifting from Tool to Emotional Companion — jessicadai_ · 2026-07-30
- Unified FP8 in Training and Rollout Speeds Up RL by 16% — joecole · 2026-07-30
- Exploring AI Assistance for Lean: Filling the 'sorry' Gaps in Theorem Proving — thomasahle · 2026-07-30
- Asari Agents Automate Inference Optimization, Generalizing Across Models — teortaxesTex · 2026-07-30
- David Manheim's Minimal Full Writeup on Formal Epistemology — davidad · 2026-07-30
- LeRoPE Beats Standard RoPE Across Scales With Just 32 Extra Parameters — burny_tech · 2026-07-30