GSCV++: Lightweight 3D Gaussian Splatting Sequence Compression with Standard Video Codecs

Forum topic · 小凯 · 2026-11-04

Summary

GSCV++ is a lightweight pipeline for compressing 3D Gaussian Splatting (GS) sequences using standard video codecs, targeting industrial needs for low complexity and hardware compatibility. The paper improves three stages jointly. First, energy-adaptive inter-frame spherical harmonic AC dimensionality reduction (EA-SHDR) produces a compact and temporally stable SH AC feature space, cutting data volume and complexity. Second, prior-based Inter-MiniPLAS (P-MiniPLAS) generates smooth 2D feature maps via prior-based weighting factor inference and small-scale remapping, producing maps better suited to video codecs than vanilla PLAS or Morton scan while keeping low computational cost. Third, content-aware generalized compression (CA-GC) applies non-uniform bit-depth splits for coordinate map generation and coding unit size selection, exploiting codec performance without normative changes. Experiments show over 20% intra- and 60% inter-prediction rate-distortion gains versus state-of-the-art methods, plus faster 2D map generation, encoding, and decoding. arXiv: 2611.05359.

论文概要

研究领域: CV 作者: Qi Yang, Yingzhan Xu, Kai Zhang, Li Zhang 发布时间: 2026-11-03 arXiv: 2611.05359

Overview

This paper presents GSCV++, a lightweight pipeline for 3D Gaussian Splatting (GS) sequence Compression with standard Video codecs. To meet the requirements of low-complexity and hardware compatibility from the industry, GSCV++ jointly improves preprocessing, 3D-to-2D mapping, and practical video compression configuration of GS compression.

Key Contributions

1. Preprocessing: EA-SHDR

Energy-adaptive inter-frame spherical harmonic (SH) AC dimensionality reduction (EA-SHDR) produces a compact and stable temporal SH AC feature space. It reduces the data volume of the following mapping and practical compression stages, lowers complexity, and eliminates redundant information.

2. 3D-to-2D Mapping: P-MiniPLAS

Prior-based Inter-MiniPLAS (P-MiniPLAS) generates smooth 2D feature maps through prior-based weighting factor inference and small-scale remapping. Compared with vanilla Parallel Linear Assignment Sorting (PLAS) and Morton scan, P-MiniPLAS ensures the generated 2D maps are more suitable for the video codec while maintaining low computational complexity.

3. Practical Compression: CA-GC

Content-aware generalized compression (CA-GC) uses a non-uniform bit-depth split for coordinate map generation, and introduces coding unit size selection for efficient compression complexity control — fully exploiting video codec performance without normative changes.

Results

Over 20% intra-prediction and 60% inter-prediction rate-distortion gains compared with state-of-the-art (SOTA) methods
Faster 2D map generation, encoding, and decoding than other SOTA methods using the same type of video codecs
Ablation studies verify the effectiveness and robustness of the proposed modules and provide insights for subsequent research