LEXI-SG: Monocular 3D Scene Graph Mapping with Room-Guided Feed-Forward Reconstruction

Researchers introduced LEXI-SG, a monocular RGB system for dense open-vocabulary 3D scene graph mapping in indoor environments.

It reconstructs rooms one at a time and uses a room-based factor graph to align them globally, avoiding sliding-window inconsistencies.

The method improves trajectory estimation, dense reconstruction, and segmentation on indoor datasets such as Habitat-Matterport 3D.

The work shows that scalable 3D scene graph mapping can be done with a single RGB camera, reducing dependence on depth sensors and LiDAR.