Efficient 4D Gaussian Stream with Low Rank Adaptation

TL;DR

This paper introduces a scalable 4D Gaussian-based method for dynamic novel view synthesis that employs low-rank adaptation to efficiently model scene changes, significantly reducing bandwidth while maintaining high quality.

Contribution

It presents a novel continual learning approach using 3D Gaussians and low-rank adaptation for dynamic scene reconstruction in video synthesis.

Findings

Reduces streaming bandwidth by 90%

Maintains high rendering quality comparable to state-of-the-art offline methods

Enables scalable, continual scene reconstruction

Abstract

Recent methods have made significant progress in synthesizing novel views with long video sequences. This paper proposes a highly scalable method for dynamic novel view synthesis with continual learning. We leverage the 3D Gaussians to represent the scene and a low-rank adaptation-based deformation model to capture the dynamic scene changes. Our method continuously reconstructs the dynamics with chunks of video frames, reduces the streaming bandwidth by $90\%$ while maintaining high rendering quality comparable to the off-line SOTA methods.