Mingrui Li and Shuhong Liu Claim SGS-SLAM Is First Semantic Gaussian SLAM
Summary
Mingrui Li, Shuhong Liu and co-authors release SGS-SLAM, which they describe as the first semantic visual SLAM system built on 3D Gaussian Splatting. The BSD-3-Clause code combines appearance, geometry and semantic features, and targets real-time rendering on Replica and ScanNet scenes.
Key Points
- A semantic feature loss offsets limits of traditional depth and color losses in object optimization, the authors write.
- A semantic-guided keyframe selection strategy stops cumulative errors from causing erroneous reconstructions, per the authors.
- The system stores reconstructions in .npz files carrying appearance and semantic features, viewable in SuperSplat.