MVFusion-GS: Motion-Variance Guided Temporal Attention for High-Quality Dynamic Gaussian Splatting

2026-07-01

AI Summary

This work addresses the limitations of existing dynamic 3D Gaussian splatting methods, which often fail to accurately model foreground motion due to the absence of explicit motion awareness and tend to introduce spurious static residuals in the background. To overcome these issues, the authors propose a motion-variance-guided dynamic-static separation mechanism and introduce MotionFormer, a Transformer-based local temporal attention module that enables explicit short- and long-term motion modeling within the Gaussian splatting framework for the first time. By integrating motion variance statistics, deformation field optimization, and self-attention mechanisms, the method significantly enhances both foreground motion accuracy and static background fidelity, achieving state-of-the-art performance on dynamic scene reconstruction and occlusion-free reconstruction benchmarks.

Abstract

3D Gaussian Splatting (3DGS) enables real-time novel view synthesis for static scenes. Extending it to dynamic scenes via deformation fields has recently attracted significant attention, particularly for dynamic scene reconstructionband distractor-free. However, existing deformation networks lack explicit motion awareness: they neither capture long-term motion intensity nor exploit short-term temporal coherence, leading to inaccurate foreground deformation and pseudo-static residuals in the background. We present MVFusion-GS, a method that enhances deformation networks with two complementary motion-aware mechanisms. The Motion-Variance Guided Refinement aggregates per-Gaussian deformation statistics across time to estimate motion variance and uses it to guide dynamic-static separation during deformation prediction. The MotionFormer Temporal Attention module applies Transformer self-attention over neighboring timesteps to model local motion dependencies and improve temporal consistency. Extensive experiments on both dynamic scene reconstruction and distractor-free reconstruction benchmarks demonstrate state-of-the-art performance, showing that explicit motion awareness improves both foreground motion modeling and static background reconstruction.

Problem

• Dynamic Gaussian Splatting
• Deformation Fields
• Motion Awareness
• Temporal Coherence
• Foreground-Background Separation

Innovation

• Motion-Variance Guidance
• Temporal Attention
• Dynamic Gaussian Splatting
• Motion-Aware Deformation
• Transformer-based Temporal Modeling