OUGS: Active View Selection via Object-aware Uncertainty Estimation in 3DGS

Recent advancements in 3D Gaussian Splatting (3DGS) have led to state-of-the-art results in novel view synthesis, yet challenges remain in capturing high-fidelity reconstructions of specific objects within complex scenes. Existing methods often rely on scene-level uncertainty metrics, which can be biased by background clutter, leading to inefficient view selection. The newly introduced OUGS framework addresses this limitation by deriving uncertainty directly from the explicit physical parameters of 3D Gaussian primitives, such as position and scale. This innovative approach allows for a more interpretable uncertainty model and integrates semantic segmentation masks to produce targeted, object-aware uncertainty scores. Experimental evaluations demonstrate significant improvements in both the efficiency of the 3DGS reconstruction process and the quality of targeted objects compared to existing methods, highlighting OUGS's potential to transform object-centric tasks in 3D reconstruction.