MrNeRF (@janusch_patas)
2026-01-27 | โค๏ธ 39 | ๐ 7
LoD-Structured 3D Gaussian Splatting for Streaming Video Reconstruction
Contributions: โข We propose an Anchor- and Octree-based LoD-structured 3DGS representation, integrated with a hierarchical Gaussian dropout technique to stabilize training and ensure high-fidelity synthesis in Sparse-View SFVV.
โข We introduce a GMM-driven mechanism to decouple dynamic and static content, enabling targeted refinement of moving regions while maintaining the structural stability of the background.
โข We develop a quantized residual refinement framework that compresses dynamic updates, facilitating efficient data transmission and storage for low-bandwidth environments.
โข Extensive experiments demonstrate that StreamLoD-GS achieves competitive or state-of-the-art performance in terms of quality, efficiency, and storage footprints compared to contemporary SFVV approaches.
๋ฏธ๋์ด
