Handling dynamic objects in 3D Gaussian Splatting (3DGS) consolidation

Develop 3D Gaussian Splatting (3DGS) techniques that robustly model and render dynamic objects during 3D consolidation, achieving consistent multi-view reconstruction and rendering in dynamic scenes rather than only static scenes.

Background

The proposed pipeline consolidates densified X-images and RGB views in a unified 3D space using 3D Gaussian Splatting (3DGS), relying on COLMAP poses for RGB. While effective for static scenes, the authors note that dynamic objects pose a challenge for 3D consolidation, which current 3DGS methods typically assume to be static.

The limitation highlights a broader research gap: extending 3DGS beyond static radiance fields to handle scene dynamics, ensuring temporal and multi-view consistency when objects move or deform. Addressing this would enable cross-sensor view synthesis and consolidation in more realistic, dynamic environments.