BROWSE
Detail
Conference Paper
Structured Gaussian Mapping for Memory-Efficient Immersive XR
Cited - time in 
Share 
Share
- Authors
- Kim Hye-Sun, Hong Sungjin, Cho-Rong Yu, Gil Youn-Hee
- Issue Date
- 2026-03
- Citation
- Conference on Virtual Reality and 3D User Interfaces Abstracts and Workshops (VRW) 2026, pp.1123-1124
- Publisher
- IEEE
- Language
- English
- Type
- Conference Paper
- Abstract
- High-fidelity and scalable 3D scene representations are critical for immersive VR and XR systems under strict latency and memory constraints. Although recent studies demonstrate real-time 3D Gaussian Splatting on head-mounted displays, uncontrolled Gaussian growth during RGB-D mapping remains a key scalability bottleneck. We propose Structured Gaussian Mapping, a KD-tree guided allocation framework that limits Gaussian growth at the mapping stage for sequential RGB-D SLAM. By partitioning image space and allocating compact Gaussian sets only in perceptually relevant regions, our approach replaces pixel-wise densification. Experiments show up to 82% fewer Gaussians with improves PSNR, enabling memory-efficient and scalable XR-ready Gaussian mapping.
- Keyword
- 3D Gaussian Splatting, RGB-D SLAM, KD-tree Partitioning, Scalable XR Mapping
- KSP Keywords
- 3D scene, Head mounted displays(HMD), High-fidelity, Immersive VR, K-d tree, RGB-D SLAM, RGB-D mapping, Real-time 3d, Tree partitioning, memory-efficient