Dropping Anchor and Spherical Harmonics for Sparse-view Gaussian Splatting

Shuangkang Fang, I-Chao Shen, Xuanyang Zhang, Zesheng Wang, Yufeng Wang, Wenrui Ding, Gang Yu, Takeo Igarashi · 2026

Read Dropping Anchor and Spherical Harmonics for Sparse-view Gaussian Splatting by Shuangkang Fang, I-Chao Shen, Xuanyang Zhang, Zesheng Wang, Yufeng Wang, Wenrui Ding, Gang Yu, Takeo Igarashi (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-2602-20933.txt.

Description

Three-dimensional Gaussian splatting overfits when only a few views are available, and existing dropout methods fight it by randomly zeroing Gaussian opacities. Shuangkang Fang and seven co-authors show a flaw in that approach in this 2026 arXiv paper: neighbors compensate for dropped Gaussians, blunting the regularization. Their method, DropAnSH-GS, picks anchor Gaussians and removes their spatial neighbors too.

Dropping anchors with their neighborhoods breaks local redundancy and pushes the model toward robust, global representations. The method also drops high-degree spherical harmonic coefficients from color, concentrating appearance in lower degrees. That further curbs overfitting and allows post-training compression by truncating harmonics. Experiments show clear gains over existing dropout methods at negligible extra cost.

Catalog record

CREATORShuangkang Fang, I-Chao Shen, Xuanyang Zhang, Zesheng Wang, Yufeng Wang, Wenrui Ding, Gang Yu, Takeo Igarashi
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE34 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-20933

Browse the full library · About the Tower of Babel