Given a posed RGBD stream, GaME reconstructs a 3D scene represented with 3D Gaussians. Our mapping system can adapt to the changes occurring in the scene that are not directly observable. For example, you reconstructed your kitchen, continued to capture your bedroom, and then returned to the kitchen. During your absence, someone moved the chair and added a table in the kitchen. Our method can handle such situations.
GaME Architecture.Given a segmented RGB-D input stream, the keyframe management system selects keyframes and triggers the dynamic scene adaptation (DSA) module. DSA first integrates newly observed geometry, then removes outdated geometry using covisible keyframes from the 3D Gaussian Splatting map. The keyframe manager then masks stale regions, and the mapping system uses the processed keyframes for local covisibility window optimization.
The Aria Dataset features fewer changes compared to the Flat Dataset. Notable differences include items on the table and a picture that was moved to the wall between runs. Another example is the tea table, which was also repositioned. In the second room, you can observe the picture that was moved from one shelf to another.
SplatAM
DG-SLAM
MonoGS
GaME (ours)
Ground-truth
SplatAM
DG-SLAM
MonoGS
GaME (ours)
Ground-truth
SplatAM
DG-SLAM
MonoGS
GaME (ours)
Ground-truth
SplatAM
DG-SLAM
MonoGS
GaME (ours)
Ground-truth
SplatAM
DG-SLAM
MonoGS
GaME (ours)
Ground-truth
SplatAM
DG-SLAM
MonoGS
GaME (ours)
Ground-truth
@misc{yugay2025gaussianmappingevolvingscenes,
title={Gaussian Mapping for Evolving Scenes},
author={Vladimir Yugay and Thies Kersten and Luca Carlone and Theo Gevers and Martin R. Oswald and Lukas Schmid},
year={2025},
eprint={2506.06909},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2506.06909},
}