基于形状先验的隐式三维建模算法研究  

Research on Implicit 3D Modeling Algorithm Based on Shape Prior

作　　者：倪天杰 何良华[1] 

机构地区：[1]同济大学计算机科学与技术系,上海

出　　处：《计算机科学与应用》2023年第11期2012-2021,共10页Computer Science and Application

摘　　要：近年来,隐式神经三维建模算法是计算机视觉领域的热门研究方向,特别是基于符号距离函数(SDF)的方法,然而目前的模型往往泛化性不强。由此问题,该文利用动态图卷积网络对点云进行形状编码,引入了形状先验假设,提出了一种结合物体个性特征和其所属类别共性信息相结合的隐式三维建模算法,提升了模型的准确性与泛化性。在ShapeNetV2数据集上,相较于现有算法取得了更好的效果,表明了本方法在三维隐式建模问题上的优越性。In recent years, implicit neural 3D modeling algorithm has become a popular research direction in the field of computer vision, especially the method based on signal distance function (SDF). However, this kind of model is often not strong generalization. To solve this problem, this paper uses the dynamic graph convolutional network to encode the shape of objects, introduces the shape prior hy-pothesis, and proposes an implicit 3D modeling algorithm that combines the individual characteristics of objects and the generic information of their categories, which improves the accuracy and generalization of the model. On ShapeNetV2 dataset, better results are obtained compared with the existing algorithms, which shows the superiority of the proposed method in three-dimensional implicit modeling.

关 键 词：隐式神经场 深度学习 三维建模 点云卷积 符号距离函数 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]