摘要
为了提高三维人体重建精度并使得重建结果更加可控,设计了一种基于图卷积的三维人体重建方法。该方法不依赖任何现有的参数化人体模型,以人体掩码图像和少量的人体测量尺寸作为输入,借助图卷积神经网络直接回归三维人体网格模型的顶点坐标,其本质是利用图卷积算子对内置的模板人体进行变形。大量实验证明,通过显式地融入人体测量数据并辅以相应的损失函数,重建精度大幅提高,重建人体的各项测量尺寸误差均小于1 cm,且重建效果优于其他相关方法。
A non-parametric 3D human body reconstruction method based on Graph Convolutional Network(GCN),which does not depend on any existing parametric human body model,was proposed in this paper to improve the precision of reconstruction and make the procedure more controllable.The proposed method only required mask image(s)and a small of anthropometric measurements of a body shape as input and regresses the 3D coordinates as output directly,whose essence was to employ the graph convolutional operator to deform the built-in body template.Experimental results demonstrate that by explicitly integrating the anthropometric sizes into the network with a properly designed loss function,the accuracy of the reconstruction is greatly improved,all anthropometric errors are less than 1 cm,and the reconstruction result is better than other related methods as well.
作者
谢昊洋
钟跃崎
XIE Haoyang;ZHONG Yueqi(College of Information Engineering,North China University of Water Resources and Electric Power,Zhengzhou,Henan 450046,China;College of Textiles,Donghua University,Shanghai 201620,China;Key Laboratory of Textile Science&Technology,Ministry of Education,Donghua University,Shanghai 201620,China)
出处
《毛纺科技》
CAS
北大核心
2021年第4期18-24,共7页
Wool Textile Journal
关键词
三维人体
重建
图卷积网络
非参数化建模
three-dimensional human body
reconstruction
Graph Convolutional Network
nonparametric modeling
