Semantic-Guided 3D Gaussian Splatting for Transient Object Removal
Aditi Prabakaran, Priyesh Shukla · 2026
Read Semantic-Guided 3D Gaussian Splatting for Transient Object Removal by Aditi Prabakaran, Priyesh Shukla (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-15516.txt.
Description
Prabakaran and Shukla tackle ghosting in 3D Gaussian Splatting, where passing people, cars, or other brief objects smear into reconstructions built from casual multi-view photos. Older fixes split the scene at high memory cost or leaned on motion cues that parallax can fool. The authors instead filter by meaning: a vision-language model scores each rendered view against text prompts naming distractor types, and scores pile up per Gaussian through training.
Gaussians that pass a tuned threshold face opacity penalties and periodic pruning, so known distractor kinds are cut no matter how they move. Tests on the RobustNeRF benchmark show steady gains over plain 3DGS on four sequences, with little extra memory and real-time rendering kept. The method suits scenes where the distractor kinds can be named in advance.
Catalog record
| CREATOR | Aditi Prabakaran, Priyesh Shukla |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 19 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-15516