SupGS-SLAM: Gaussian Splatting SLAM with Efficient Keyframe Strategy and Supplementary Mapping
Shuai Liu, Yongcai Wang, Wenping Chen, Wang Chen, Deying Li
AI summary
Problem
Existing Gaussian Splatting SLAM methods degrade in rendering quality for previously observed regions as the camera moves away, primarily due to redundant keyframe selection and insufficient optimization of critical areas.
Approach
The method introduces a view-overlap-based keyframe strategy that dynamically weights and prioritizes critical frames, alongside a supplementary mapping module that adds missing Gaussian primitives, optimizes the global scene, and fills depth gaps using estimated depth.
Key results
- Reduces redundant keyframes while ensuring comprehensive scene coverage
- Significantly improves full-scene rendering quality on synthetic and real-world datasets
- Maintains accurate camera tracking and mapping performance
- Effectively reconstructs regions lacking ground-truth depth using estimated depth
Why it matters
Provides a robust, high-fidelity reconstruction pipeline essential for real-time embodied AI, VR, and AR applications requiring consistent global scene quality.
Abstract
Gaussian Splatting SLAM methods have exhibited impressive high-fidelity rendering performance. Existing meth- ods maintain high rendering quality around the current camera viewpoint, but the rendering quality degrades in previously observed regions as the camera moves away, particularly in real-world scenarios. We identify two core factors for high- quality rendering: keyframes should efficiently cover the entire scene while minimizing redundancy, and the mapping strategy should effectively select critical keyframes for full scene opti- mization. To address these issues, we propose SupGS-SLAM to improve rendering quality across the entire scene. For effec- tive keyframe management, we propose an efficient keyframe strategy, which reduces redundant keyframe selection and pri- oritizes the optimization of critical keyframes by assigning high weights. For enhanced mapping, we propose a supplementary mapping strategy comprising three components: supplementary densification, supplementary global mapping, and supplemen- tary depth mapping. In supplementary densification, we add supplementary Gaussian primitives to previous regions with insufficient representation. In supplementary global mapping, we select keyframes globally to optimize the full scene. In sup- plementary depth mapping, we use estimated depth to optimize regions without ground-truth depth. Extensive experiments demonstrate that SupGS-SLAM achieves excellent performance on both synthetic and real-world datasets. The project page is available at https://github.com/rucliushuai/SupGS-SLAM.