Integrating cloth material priors into 3D Gaussian Splatting

Integrate the material priors learned from planar SVBRDF inversion effectively within a scene-level 3D Gaussian Splatting reconstruction.

Background

RoboCloth learns material priors for reconstructing spatially varying reflectance of near-planar cloth samples through a two-stage neural BRDF pipeline. The reported experiments focus on planar material capture and reconstruction, rather than jointly reconstructing complete three-dimensional scenes.

The paper identifies the extension of these learned material priors to scene-level 3D Gaussian Splatting as an unresolved research direction. Solving this problem would connect cloth-specific SVBRDF inversion with 3D scene representations and enable material-aware reconstruction beyond isolated planar samples.