ScholarWorks@성균관대학교

초록

As the demand for high-fidelity six degrees of freedom (6DoF) content grows, 3D Gaussian Splatting (3DGS) has emerged as a promising representation due to its superior rendering quality and speed. However, the massive storage requirements of 3DGS pose significant challenges for efficient transmission and web-based streaming. While the MPEG geometry-based point cloud compression (G-PCC) standard offers efficient compression for point clouds, its direct application to 3DGS is hindered by data type mismatches and the lack of a standardized container. This paper proposes a standard-compliant streaming framework that encapsulates G-PCC encoded 3DGS bitstreams within the glTF format. This paper introduce a quantization-aware encoding scheme and a custom glTF extension MPEG-GSC-EXT to preserve the float-precision attributes of Gaussians. Furthermore, we implement a client-side rendering pipeline capable of on-the-fly de-quantization. Experimental results show the proposed method achieves significant file size reduction compared to raw PLY and compressed formats while maintaining visual fidelity.

키워드