EKF-GS: An Improved 3D Gaussian Splatting Using Extended Kalman Filter
We propose EKF-GS, a hybrid optimization framework for 3D Gaussian Splatting that integrates Extended Kalman Filter with stochastic gradient descent. EKF-GS enables faster convergence, uncertainty-guided Gaussian densification, and uncertainty quantification capability. Experiments on public datasets show improved reconstruction quality with fewer iterations and reduced training time.
Authors
- Kezhi Wang (ORCID: https://orcid.org/0000-0001-8602-0800)
- Zhenyu Chen (ORCID: https://orcid.org/0000-0002-4989-7109)
- Jikai Wang
Institutions
- University of Science and Technology of China (CN)
Publication Details
- Journal
- IEEE Transactions on Visualization and Computer Graphics
- Published
- 2025-12-22
- DOI
- https://doi.org/10.1109/tvcg.2025.3646761
- Primary Topic
- Robotics and Sensor-Based Localization
- Type
- article
- Field-Weighted Citation Impact
- 0.00