Differential Dynamic Gaussian Splatting: Full Scene Scalable Volumetric Video Reconstruction
Loading...
Date
2025
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
The Eurographics Association
Abstract
The growing demand for immersive experiences in AR/VR, sports, and cinematic content is accelerating interest in volumetric video. However, existing dynamic extensions of 3D Gaussian Splatting (3DGS) methods struggle with large memory footprints, limited scalability, and impractical storage requirements-especially for long sequences on consumer-grade hardware. We present a scalable, resource-efficient pipeline for full-scene Dynamic Gaussian Splatting, achieving real-time volumetric video reconstruction on consumer-grade GPUs with <1MB per frame storage and <7.6GB memory usage, all while preserving high visual fidelity and seamless handling of emerging objects. Our method combines batch-wise differential training with per-frame PLY-based storage, enabling arbitrarily long sequences with dynamic content. By leveraging standard file formats and lightweight computation, our system lowers the barrier to entry for researchers and developers aiming to integrate high-quality volumetric video into AR/VR and interactive graphics toolchains.
Description
CCS Concepts: Computing methodologies → Rendering; Volumetric models; Virtual reality
@inproceedings{10.2312:cgvc.20251205,
booktitle = {Computer Graphics and Visual Computing (CGVC)},
editor = {Sheng, Yun and Slingsby, Aidan},
title = {{Differential Dynamic Gaussian Splatting: Full Scene Scalable Volumetric Video Reconstruction}},
author = {Hirt, Felix and Lu, Tianxiang},
year = {2025},
publisher = {The Eurographics Association},
ISBN = {978-3-03868-293-6},
DOI = {10.2312/cgvc.20251205}
}