Reproduce “Deformable Triangle Splatting: Flexible Primitives for Real-Time Radiance Field Rendering” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Oriol Jiménez-Ayguadé, Antonio Agudo
DETRIS extends triangle primitives with learnable control points along each edge, allowing outlines to expand, contract, and form concavities while preserving a triangle’s plane. It uses barycentric coordinates and differentiable distance-based rasterization. On nine Mip-NeRF 360 scenes, it achieves LPIPS 0.183 versus 0.191 for Triangle Splatting, but trains more slowly and does not demonstrate fewer real-scene primitives.
Key insights
- Performing all boundary deformations and distance computations within a canonical barycentric coordinate space ensures strict view-independence of the primitive shape, leading to consistent appearance and stable multi-view training.
- A combination of a non-zero winding number rule and a C1-continuous polynomial smooth minimum provides a robust and differentiable mechanism for handling inside/outside classification and distance field generation for complex, potentially non-convex or self-intersecting, deformable primitives.
- Learnable boundary deformations significantly enhance the expressiveness of individual primitives, allowing them to efficiently represent a wider variety of shapes, including non-convex forms, reducing the overall primitive count needed for complex scene modeling.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.