Failing to Reflect Reality: Gaussian Splatting’s Collapse on Non-Lambertian Surfaces

IF 2.6 Q2 ENGINEERING, ELECTRICAL & ELECTRONIC IEEE open journal of signal processing Pub Date : 2026-01-01 Epub Date: 2026-04-21 DOI:10.1109/OJSP.2026.3685574
Eva Dubar;Sarah Dury;Gauthier Lafruit;Daniele Bonatto
{"title":"Failing to Reflect Reality: Gaussian Splatting’s Collapse on Non-Lambertian Surfaces","authors":"Eva Dubar;Sarah Dury;Gauthier Lafruit;Daniele Bonatto","doi":"10.1109/OJSP.2026.3685574","DOIUrl":null,"url":null,"abstract":"We explore the applicability of Gaussian Splatting (GS) to the view synthesis (VS) of non-Lambertian objects. Materials with anisotropic appearance have always been a failure case for VS, until the recent introduction of spherical harmonics promising satisfying results to handle view-directional aspects. Nevertheless, we demonstrate that highly non-Lambertian objects, such as glass or mirrors, remain impossible to correctly reconstruct and render with Gaussian primitives and spherical harmonics. This paper clarifies the reasons: (1) Gaussian Splatting optimization converges to incorrect geometry, and (2) spherical harmonics fail to represent consistent view dependencies. Benchmarking experiments confirm that gs produces false geometry even with ground truth data. Our findings call into question current material representations in view synthesis and motivate the use of more realistic primitives.","PeriodicalId":73300,"journal":{"name":"IEEE open journal of signal processing","volume":"7 ","pages":"447-455"},"PeriodicalIF":2.6000,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11489163","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"IEEE open journal of signal processing","FirstCategoryId":"1085","ListUrlMain":"https://ieeexplore.ieee.org/document/11489163/","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"2026/4/21 0:00:00","PubModel":"Epub","JCR":"Q2","JCRName":"ENGINEERING, ELECTRICAL & ELECTRONIC","Score":null,"Total":0}
引用次数: 0

Abstract

We explore the applicability of Gaussian Splatting (GS) to the view synthesis (VS) of non-Lambertian objects. Materials with anisotropic appearance have always been a failure case for VS, until the recent introduction of spherical harmonics promising satisfying results to handle view-directional aspects. Nevertheless, we demonstrate that highly non-Lambertian objects, such as glass or mirrors, remain impossible to correctly reconstruct and render with Gaussian primitives and spherical harmonics. This paper clarifies the reasons: (1) Gaussian Splatting optimization converges to incorrect geometry, and (2) spherical harmonics fail to represent consistent view dependencies. Benchmarking experiments confirm that gs produces false geometry even with ground truth data. Our findings call into question current material representations in view synthesis and motivate the use of more realistic primitives.
查看原文
分享 分享
微信好友 朋友圈 QQ好友 复制链接
本刊更多论文
不能反映现实:高斯飞溅在非朗伯曲面上的坍缩
探讨了高斯溅射(GS)在非朗伯物体视图合成(VS)中的适用性。具有各向异性外观的材料一直是VS的失败案例,直到最近引入的球面谐波有望在处理视图方向方面取得令人满意的结果。然而,我们证明了高度非朗伯物体,如玻璃或镜子,仍然不可能正确地重建和渲染高斯原语和球面谐波。本文阐明了原因:(1)高斯飞溅优化收敛于不正确的几何;(2)球面谐波不能表示一致的视图依赖关系。基准测试实验证实,即使使用地面真值数据,gs也会产生错误的几何形状。我们的发现对当前视图合成中的材料表示提出了质疑,并激发了更现实的原语的使用。
本文章由计算机程序翻译,如有差异,请以英文原文为准。
求助全文
约1分钟内获得全文 去求助
来源期刊
CiteScore
5.30
自引率
0.00%
发文量
0
审稿时长
22 weeks
期刊最新文献
Robust Kalman Filtering via Correntropy-Based Higher-Order Moment Adaptation and Variable Structure Gains Diverse Subset Selection via Norm-Based Sampling and Orthogonality Robust Localization in Modern Cellular Networks Using Global Map Features Alternatives to Sine Carrier in Auditory BCI: Exploring Machine Learning Strategies for Assessing Modulation Detectability in EEG Sampling Method for Generalized Graph Signals With Pre-Selected Vertices via DC Optimization
wanfang
×
引用
GB/T 7714-2015
复制
MLA
复制
APA
复制
导出至
BibTeX EndNote RefMan NoteFirst NoteExpress
×
×
提示
您的信息不完整,为了账户安全,请先补充。
现在去补充
×
提示
您因"违规操作"
具体请查看互助需知
我知道了
×
提示
现在去查看 取消
×
提示
确定
0
微信
客服QQ
Book学术公众号 扫码关注我们
反馈
×
意见反馈
请填写您的意见或建议
请填写您的手机或邮箱
已复制链接
已复制链接
快去分享给好友吧!
我知道了
×
扫码分享
扫码分享
Book学术官方微信
Book学术官方微信
Book学术文献互助
Book学术文献互助群
群 号:604180095
Book学术
文献互助 智能选刊 最新文献 互助须知 联系我们:info@booksci.cn
Book学术提供免费学术资源搜索服务,方便国内外学者检索中英文文献。致力于提供最便捷和优质的服务体验。
Copyright © 2023 Book学术 All rights reserved.
ghs 京公网安备 11010802042870号 京ICP备2023020795号-1