网格辅助的调色板NeRF外观编辑模型  

Palette NeRF appearance editing model with mesh assistance

作　　者：毕成龙 BI Chenglong(School of Computer Science and Engineering,Anhui University of Science and Technology,Huainan 232001,China)

机构地区：[1]安徽理工大学计算机科学与工程学院,安徽淮南232001

出　　处：《哈尔滨商业大学学报(自然科学版)》2025年第1期18-24,共7页Journal of Harbin University of Commerce:Natural Sciences Edition

基　　金：安徽省自然科学基金(2308085MF218);安徽省高等学校科学研究项目(2022AH040113)。

摘　　要：目前基于调色板的神经辐射场(Neural Radiance Field,NeRF)模型虽然能够对其生成的3D场景进行直观且细粒度的外观编辑,但无法实时渲染出高质量的视图,且局部外观编辑的细粒度被限制在目标级物体.针对这些问题,提出一种网格辅助的调色板NeRF外观编辑模型.该模型通过将场景的颜色分解为高光颜色和由调色板颜色Alpha混合形成的漫反射颜色来实现外观编辑,并将颜色烘焙到场景的三角网格模型表面以加速渲染.局部编辑时,场景的网格模型被用来辅助编辑目标的选择.实验结果表明,在不同数据类型的公共数据集下能够对任意大小和形状的3D区域进行外观编辑,并实时渲染出高质量的编辑结果.该模型使用网格加速了调色板NeRF的渲染速度并实现了像素级的局部外观编辑.The current palette-based Neural Radiance Field(NeRF)models enabled intuitive and fine-grained appearance editing of the 3D scenes.However,the fine-grained local appearance editing was limited to target-level objects,and could not render high-quality views in real time.To address these issues,a palette NeRF appearance editing model with mesh-assisted was proposed.The model achieved appearance editing by decomposing the colors of the scene into highlight colors and diffuse colors formed by alpha blending of palette colors and baking the colors onto the surface of the scene s mesh model to speed up rendering.During local editing,the mesh model of the scene was used to assist in the selection of editing targets.The experimental results showed that the model was able to edit the appearance of 3D regions of arbitrary size and shape and render high-quality editing results in real time under public datasets of different data types.The model used meshes to accelerate the rendering of the palette NeRF and enabled pixel-level local appearance editing.

关 键 词：调色板 神经辐射场 外观编辑 实时渲染 局部编辑 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]