Reproduce “MotionGS-SLAM: Event-Modulated Gaussian Splatting for Motion-Blur Robust SLAM” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Zhiqiang Hu, Shouren Huang, Masatoshi Ishikawa
MotionGS-SLAM introduces an event-modulated 3D Gaussian Splatting framework that reformulates motion blur as a generative forward-modeling task to achieve robust Simultaneous Localization and Mapping. The system produces sharp, high-fidelity 3D scene reconstructions and accurate camera tracking, reducing Absolute Trajectory Error by 51.9% compared to MonoGS on synthetic datasets, despite severe motion blur in input images.
Key insights
- Reformulating motion blur as a generative forward-modeling problem, by physically simulating the blur formation, offers greater robustness than inverse deblurring, especially under severe blur conditions.
- Event cameras provide crucial, blur-immune motion information that can be effectively integrated to guide the differentiable rendering of 3D Gaussians and improve real-time SLAM performance.
- Dynamically modulating the 3D Gaussian kernel both spatially (anisotropic, motion-aligned elongation) and temporally (adaptive exposure integral sampling) captures the physical effects of motion blur, leading to higher-fidelity scene reconstruction and more accurate pose estimation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.