摘要
表面网格重建和纹理恢复在促进神经辐射场(NeRF)广泛应用方面具有重要作用,然而,现有方法往往受限于特定的NeRF模型,并且相比隐式表达方法降低了渲染质量。为此,提出一种面向NeRF的通用纹理网格恢复方法。首先,采用等值面提取算法从NeRF模型中提取粗糙的表面网格;接着利用可微渲染器和一个颜色分离的外观模型,从多视角图像中优化表面网格并分离漫反射颜色和高光颜色;然后将漫反射颜色和高光颜色烘焙成纹理图像,并在现有渲染框架中进行运行。实验表明该方法能够提升表面网格质量,并实现具有竞争力的新视图渲染质量。
Surface mesh reconstruction and texture restoration play an important role in promoting the wide application of neural radiation field(NeRF).However,existing methods are often limited to specific NeRF models and reduce rendering quality compared to implicit representation methods.Therefore,we proposed a general mesh texture restoration method for NeRF in this paper.Firstly,we used marching cubes algorithm to extract rough surface mesh from NeRF model.Then,we used a renderer and a color separation appearance model to optimize the surface mesh from multi-view images and separate diffuse and highlight colors.Finally,we baked the diffuse and highlight colors into a textured image and ran in an existing rendering framework.Experiments show that the proposed method can improve the surface mesh quality.
作者
孙聪
张彦波
陈晏涛
王伍龙
张兵堂
李刚
SUN Cong;ZHANG Yanbo;CHEN Yantao;WANG Wulong;ZHANG Bingtang;LI Gang(Wuhan Ecological Environment Design Research Institute Co.,Ltd.,Wuhan 430050,China;Wuhan Urban Flood Control Survey and Design Institute Co.,Ltd.,Wuhan 430010,China;School of Remote Sensing and Information Engineering,Wuhan University,Wuhan 430079,China)
出处
《地理空间信息》
2026年第2期6-9,共4页
Geospatial Information
关键词
三维重建
新视图合成
表面网格
纹理提取
3D reconstruction
new view synthesis
surface mesh
texture extraction
