Determine Whether Learned Textures Can Be Safely Filtered for Level-of-Detail

Determine whether per-primitive learned textures that augment Gaussian primitives in primitive-based volumetric representations can be safely prefiltered to support level-of-detail, i.e., establish whether such simple learned textures can be filtered without violating appearance or causing artifacts when used for level-of-detail control.

Background

The paper surveys level-of-detail (LOD) techniques for volumetric rendering and discusses recent approaches that enhance Gaussian primitives with learned textures to increase modeling power per primitive. While these texturing alternatives offer compression and expressiveness, their compatibility with standard LOD practices—particularly prefiltering across scales—has not been established.

In contrast, the authors’ proposed Gabor Fields provide a frequency- and orientation-selective decomposition that enables natural LOD by masking primitives rather than storing separate prefiltered hierarchies. This highlights an unresolved question about whether learned textures attached to Gaussian primitives can similarly be filtered safely for LOD.