Representing fuzzy and strand-like geometry such as hair, fur and other thin, directionally structured materials remain a challenge for surface-based neural rendering methods. Recent approaches provide accurate surface reconstructions but rely on isotropic parameterizations that struggle to capture directionally varying properties. We explore whether introducing directional information into these parameterizations can better model such geometry. Motivated by anisotropic kernels in 3D Gaussian Splatting, we propose two extensions that augment the kernel with either viewing direction or local surface normals. We integrate these variants into surface-based rendering pipeline and perform an evaluation on selected $\textit{Shelly}$ scenes. Our findings indicate that incorporating directional information improves reconstruction quality, with normal-aligned kernels performing best on strand-like fur and view-aligned kernels excelling on highly anisotropic geometry. While our study is limited to two scenes, the results suggest that anisotropic parameterizations merit further investigation for modeling fuzzy materials.
PDF