Generalization of RT-GS2 Beyond Indoor Environments

Determine whether RT-GS2, a generalizable semantic feature learning approach for 3D Gaussian Splats that trains a set of networks on a small dataset of scenes, can generalize beyond indoor environments.

Background

Within the discussion of prior segmentation methods for 3D Gaussian Splats, the paper reviews RT-GS2 as the only generalizable technique involving semantic feature learning trained on a small dataset of scenes. While noting its promise, the authors highlight uncertainty about its applicability outside indoor settings.

This uncertainty is presented alongside the observation that RT-GS2 reports only 2D novel view segmentation results. Establishing whether RT-GS2 extends to outdoor or more diverse environments would clarify its practical scope for broader 3DGS segmentation tasks.

References

Many of the same limitations apply to RT-GS2 , the only generalizable technique involving semantic feature learning, which trains a set of networks on a small dataset of scenes, but it is not clear if this work can generalize beyond indoor environments, and only 2D novel view segmentation results are presented.

— ArtisanGS: Interactive Tools for Gaussian Splat Selection with AI and Human in the Loop  (2602.10173 - Tsang et al., 10 Feb 2026) in Section 2.2 (Segmenting 3D Gaussian Splats)

Our 3D evaluation is restricted to ScanNet indoor scenes; generalization to building-scale, outdoor, or dynamic scenes remains to be verified.

— NormLift: From Lifted Features To Semantic Reliability In 3D Gaussian Splatting  (2609.18898 - Zang et al., 16 Sep 2026) in Limitations, Section 5 (Conclusion)

Although they can be applied to our task, their behavior in large-scale city scenes remains unclear, due to the limited variation in object scale and narrow navigation space of their datasets.

— Multi-Scale Semantic Mapping in Urban Environments via Observation Calibration and Policy Dependence Regularization  (2609.34833 - Long et al., 28 Sep 2026) in Section 1, Introduction