跳到主要导航 跳到搜索 跳到主要内容

FGS-SLAM: A Calibration-Free Slam Framework Using Feed-Forward 3D Gaussian Splatting

  • Xuefeng Yang*
  • , Tongtai Cao
  • , Yue Liu
  • *此作品的通讯作者
  • Beijing Institute of Technology

科研成果: 书/报告/会议事项章节会议稿件同行评审

摘要

3D Gaussian Splatting (3DGS)-based Simultaneous Localization and Mapping (SLAM) has recently emerged as a promising paradigm for AR/VR, enabling real-time tracking and highfidelity photorealistic rendering. However, existing methods typically require accurate camera intrinsic calibration and suffer from slow 3DGS map construction, limiting immersive experiences. We present a plug-and-play monocular 3DGS-SLAM system that employs a modular feed-forward model to infer 3D Gaussian parameters from two uncalibrated views, together with a map refinement strategy tailored for feed-forward Gaussians. This is the first monocular SLAM framework leveraging a two-view 3DGS reconstruction prior, achieving real-time performance at 20 FPS.

源语言英语
主期刊名Proceedings - 2026 IEEE Conference on Virtual Reality and 3D User Interfaces Abstracts and Workshops, VRW 2026
出版商Institute of Electrical and Electronics Engineers Inc.
1247-1248
页数2
ISBN(电子版)9798319505293
DOI
出版状态已出版 - 2026
活动2026 IEEE Conference on Virtual Reality and 3D User Interfaces Abstracts and Workshops, VRW 2026 - Daegu, 韩国
期限: 21 3月 202625 3月 2026

出版系列

姓名Proceedings - 2026 IEEE Conference on Virtual Reality and 3D User Interfaces Abstracts and Workshops, VRW 2026

会议

会议2026 IEEE Conference on Virtual Reality and 3D User Interfaces Abstracts and Workshops, VRW 2026
国家/地区韩国
Daegu
时期21/03/2625/03/26

指纹

探究 'FGS-SLAM: A Calibration-Free Slam Framework Using Feed-Forward 3D Gaussian Splatting' 的科研主题。它们共同构成独一无二的指纹。

引用此