基于深度学习的三维物体重建方法研究综述

A review of 3D objects reconstruction based on deep learning

从图像、视频和点云等有限条件的输入中重建三维物体,是目前计算机视觉和计算机图形学领域重要的研究方向,是实现元宇宙的重要技术基础。在对基于深度学习三维物体重建方法的综述中,首先分析了基于多视图几何的三维物体重建方法;从三维物体的点云、体素、隐域场和网格等重建结果的角度,详细分析了基于深度学习的三维重建方法;简单讨论了深度学习和多视图几何理论相结合的三维物体重建方法。其次,对用于深度学习的三维物体重建中的损失函数、网络架构和相关数据集进行了讨论。最后,凝练了基于深度学习的三维物体重建之研究趋势。

Reconstruction of 3D objects from limited inputs such as images,videos and point clouds is currently an important research topic in the fields of computer vision and computer graphics,and an important foundation for the realization of the meta universe.In the review of 3D object reconstruction methods based on deep learning,the 3D object reconstruction methods based on multi-view geometry are analyzed first.From the perspective of the reconstruction results of point cloud,voxel,hidden field and grid of 3D objects,the 3D reconstruction method based on deep learning is analyzed in detail,and the 3D object reconstruction method based on deep learning and multi-view geometry theory is briefly discussed.Secondly,the loss function,network architecture and related data sets in 3D object reconstruction for deep learning are discussed.Finally,the research trend of3D object reconstruction based on deep learning is condensed.