Compact 3D Scene Representation via Self-Organizing Gaussian Grids

• DOI: 10.48550/arxiv.2312.13299
• Publication Date: 2023-01-01
• Authors: Wieland Morgenstern; Florian Barthel; Anna Hilsmann; Peter Eisert
• Keywords: FOS: Computer and information sciences; Computer Vision and Pattern Recognition (cs.CV); Computer Science - Computer Vision and Pattern Recognition

3D Gaussian Splatting has recently emerged as a highly promising technique for modeling of static scenes. In contrast to Neural Radiance Fields, it utilizes efficient rasterization allowing very fast rendering at high-quality. However, the storage size is significantly higher, which hinders practical deployment, e.g. on resource constrained devices. this paper, we introduce compact scene representation organizing parameters (3DGS) into 2D grid with local homogeneity, ensuring drastic reduction in requirements without compromising visual quality during rendering. Central our idea explicit exploitation perceptual redundancies present natural essence, inherent nature allows numerous permutations equivalently represent it. To end, propose novel parallel algorithm that regularly arranges high-dimensional while preserving their neighborhood structure. During training, further enforce smoothness between sorted grid. The uncompressed Gaussians use same structure 3DGS, seamless integration established renderers. Our method achieves factor 17x 42x complex scenes no increase training time, marking substantial leap forward domain distribution and consumption. Additional information can be found project page: https://fraunhoferhhi.github.io/Self-Organizing-Gaussians/