Compressing 3D Gaussian Splatting (3DGS) models is an active area of research. A new arXiv paper describes a learned compression method that is compatible with standard video codecs. The approach follows a video-based route: it represents Gaussian attributes as 2D maps and codes them with those codecs.
According to the abstract, the work focuses on high-dimensional non-DC spherical attributes among these maps. The abstract is truncated at that point, so further details are not available in the announcement. The paper's contribution is a codec-compatible learned compression scheme, which could make 3DGS compression easier to deploy with existing video infrastructure.
The source is an arXiv cross announcement, so no experimental results or comparisons are included in the available text.