北京大学学报（自然科学版） ›› 2025, Vol. 61 ›› Issue (4): 733-745.DOI: 10.13209/j.0479-8023.2024.095

上一篇    下一篇

基于野外全景视频数据的3D Gaussian Splatting渲染质量评估方法

姜金廷1, 陈斌2,†   

  1. 1. 北京大学地球与空间科学学院, 北京 100871 2. 北京大学计算机学院, 北京 100871
  • 收稿日期:2024-04-18 修回日期:2024-05-13 出版日期:2025-07-20 发布日期:2025-07-20
  • 通讯作者: 陈斌, E-mail: gischen(at)pku.edu.cn
  • 基金资助:
    国家重点研发计划(2021YFF0500900)资助

3D Gaussian Splatting Rendering Quality Estimation Based on Outdoor Panoramic Video Data

JIANG Jinting1, CHEN Bin2,†   

  1. 1. School of Earth and Space Sciences, Peking University, Beijing 100871 2. School of Computer Science, Peking University, Beijing 100871
  • Received:2024-04-18 Revised:2024-05-13 Online:2025-07-20 Published:2025-07-20
  • Contact: CHEN Bin, E-mail: gischen(at)pku.edu.cn

PDF

摘要：

提出一种与场景高斯椭球体分布相关的渲染质量指数计算方法。该方法利用场景的全景渲染图像计算渲染质量指数, 定量地评价全景图像的渲染质量, 并利用全景图像的FID值检验渲染图像与真实图像的语义一致性。实验结果表明, 该渲染质量指数可以有效地评估全景图像渲染质量, 符合人类的直觉认知, 并获得更高的计算性能, 可为保持虚拟野外考察过程中用户的交互沉浸感提供支持。

关键词: 野外环境, 全景影像, 渲染质量, 三维重建, 场景理解

Abstract:

A rendering quality index calculation method related to the Gaussian ellipsoid distribution of the scene is proposed. Using the panoramic rendering image of the scene, the rendering quality index is calculated to quantitatively evaluate the rendering quality of the panoramic image, and test the semantic consistency between the rendered image and the real image using the FID (fréchet inception distance) of the panoramic image. The experimental results show that this rendering quality index can effectively evaluate the rendering quality of panoramic images, conform to human intuitive cognition, and obtain higher computational performance, which can provide support for maintaining user interaction immersion during virtual field trips.

Key words: outdoor environment, panoramic image, rendering quality, 3D reconstruction, scene understanding

版权所有 © 《北京大学学报（自然科学版）》编辑部
技术支持：北京玛格泰克科技发展有限公司