DSD-GS: Dynamic-Static Decomposition of Gaussian Splatting for Efficient and High-Fidelity Dynamic Scene Reconstruction

2026-05-29

AI Summary

High-fidelity dynamic scene reconstruction often suffers from high computational costs, lengthy training times, and limited reconstruction quality. This work proposes a static-dynamic decomposition framework based on Gaussian splatting that models only the dynamic regions, thereby avoiding redundant computation on static components. The method integrates a feedforward Gaussian encoder with an optical flow model and introduces an efficient deterministic initialization strategy that eliminates the need for COLMAP preprocessing. Evaluated on the Neural 3D dataset, the approach achieves rendering speeds exceeding 700 FPS after just 10 minutes of training on a single RTX 5090 GPU, significantly outperforming existing methods in training speed, rendering frame rate, and storage efficiency.

Abstract

Dynamic scene reconstruction and novel view synthesis are fundamental to next-generation visual intelligence applications such as virtual reality, robotics, and digital twins. However, high-fidelity reconstruction of complex, time-varying scenes from arbitrary viewpoints remains a significant challenge. Existing dynamic 3DGS methods suffer from computational inefficiency, since they model all Gaussians as dynamic components. While recent decomposition-based approaches address this issue, they still struggle with degraded reconstruction quality and prolonged training time. To mitigate these limitations, we propose a novel dynamic reconstruction framework built upon an efficient static-dynamic decomposition strategy using a Feed-Forward Gaussian Splatting encoder and an optical flow model. By eliminating redundant computations on static regions, our method achieves state-of-the-art performance, outperforming existing baselines across rendering quality, training and rendering speed, and storage efficiency. Notably, on the Neural 3D dataset, our framework requires only 10 minutes for training and achieves a rendering speed of over 700 FPS on a single NVIDIA RTX 5090 GPU at resolution of 1352x1014. Furthermore, our decomposition strategy eliminates the need for COLMAP preprocessing and enables deterministic initialization, thereby enhancing both efficiency and reproducibility.

Problem

• dynamic scene reconstruction
• novel view synthesis
• Gaussian Splatting
• static-dynamic decomposition
• computational efficiency

Innovation

• Gaussian Splatting
• Dynamic-Static Decomposition
• Novel View Synthesis
• Optical Flow
• Efficient 3D Reconstruction