LabelGS: Label-Aware 3D Gaussian Splatting for 3D Scene Segmentation

Yupeng Zhang, Dezhi Zheng, Ping Lu, Han Zhang, Lei Wang, Liping Xiang, Cheng Luo, Kaijun Deng, Xiaowen Fu, Linlin Shen (🖂), Jinbao Wang · 2025

Read LabelGS: Label-Aware 3D Gaussian Splatting for 3D Scene Segmentation by Yupeng Zhang, Dezhi Zheng, Ping Lu, Han Zhang, Lei Wang, Liping Xiang, Cheng Luo, Kaijun Deng, Xiaowen Fu, Linlin Shen (🖂), Jinbao Wang (2025) 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-2508-19699.txt.

Description

Yupeng Zhang and co-authors taught a fast 3D scene renderer to also say what everything in the scene is. Three-dimensional Gaussian Splatting builds scenes from millions of soft 3D blobs and renders them quickly and beautifully, but it cannot segment objects. Their method, LabelGS, attaches an object label to each blob, guided by labeled images from several viewpoints.

Three added components keep labels consistent: one handles places where objects hide each other, one lifts the main labels from images into 3D, and one filters out blobs with conflicting labels. A random region sampling scheme speeds the whole process. LabelGS beat the leading earlier method on segmentation quality while training about twenty-two times faster at high resolution.

Catalog record

CREATORYupeng Zhang, Dezhi Zheng, Ping Lu, Han Zhang, Lei Wang, Liping Xiang, Cheng Luo, Kaijun Deng, Xiaowen Fu, Linlin Shen (🖂), Jinbao Wang
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE22 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2508-19699

Browse the full library · About the Tower of Babel