ETRI-Knowledge Sharing Plaform

KOREAN
논문 검색
Type SCI
Year ~ Keyword

Detail

Conference Paper Spherical Harmonics Coefficient Prediction for Gaussian Splat Coding
Cited - time in scopus Share share facebook twitter linkedin kakaostory
Authors
Byung-Yoon Choi, Dong-Ha Kim, Kwan-Jung Oh, Jae-Gon Kim
Issue Date
2026-03
Citation
Conference on Virtual Reality and 3D User Interfaces Abstracts and Workshops (VRW) 2026, pp.162-165
Publisher
IEEE
Language
English
Type
Conference Paper
DOI
https://dx.doi.org/10.1109/VRW70859.2026.00035
Abstract
3D Gaussian Splatting (3DGS) enables high-quality free-viewpoint rendering but suffers from large data sizes due to its high-dimensional attributes, particularly spherical harmonics (SH) coefficients. In MPEG Gaussian Splat Coding (GSC), both the video-based TMCV framework and fast-track VPCC extensions are under developed; however, explicit tools that effectively exploit the structural correlation among SH coefficients remain limited. This paper proposes a prediction-based residual coding method that compresses prediction residuals as 2D attribute maps using HEVC by exploiting the correlation among SH coefficients within each color channel. The proposed method is implemented in both the Test Model Candidate Video (TMCV) and the V-PCC reference software (TMC2), and its performance is evaluated under the GSC common test conditions (CTC). For forward-facing sequences, the proposed method achieves average BD-rate reductions of 16.6% and 3.5% (in terms of RGBPSNR) compared to the TMCV and V-PCC anchors, respectively. In contrast, average BD-rate increases are observed for object-centric sequences, motivating future work on adaptive application of the proposed SH coefficient prediction method depending on sequence characteristics.
Keyword
3D Gaussian Splatting (3DGS), Gaussian Splat Coding (GSC), Spherical Harmonics (SH) coefficients, TMCV, V-PCC
KSP Keywords
Adaptive applications, Coding method, High-quality, Object-centric, Prediction methods, Prediction-based, Structural correlation, color channel, free-viewpoint rendering, high-dimensional, large data