基于深度学习的单幅图像三维重建

Single Image 3D Reconstruction Based on Deep Learning

随着深度学习技术的发展,深度神经网络在图像处理和三维重建中得到广泛应用,为探究目前深度学习框架下的单幅图像三维重建研究现状,该文对近年的相关研究工作进行综述.首先介绍深度学习框架下基于图像的不同三维重建方法的分类;其次梳理图像三维重建中不同神经网络方法的研究进展;并根据重建三维模型表示方式的不同,分别讨论针对体素、点云、网格、隐式等不同表示方式的单幅图像三维重建网络和方法;然后给出单幅图像三维重建中的常用评价指标与数据集,并对公开数据集下针对不同表示方式的各类三维重建方法的结果进行比较与分析;最后对单幅图像三维重建所面临的困难和挑战进行讨论,并给出未来的研究方向.

With the development of deep learning technology,deep neural network has been widely applied in image processing and 3D reconstruction.In order to explore the research development of single image-based 3D reconstruction under the framework of deep learning,this paper summarizes the relevant research work in recent years.Firstly,the classification of different image-based 3D reconstruction methods under the framework of deep learning was introduced.Secondly,the research progress of different neural networks for image-based 3D reconstruction based on deep learning was reviewed.Furthermore,according to the different representations of the reconstructed 3D models,the 3D reconstruction networks and methods of single image for different representations,such as voxel,point cloud,mesh,implicit representation,were discussed respectively.Then,the commonly used metrics and datasets for single image-based 3D reconstruction were given,and the results of various reconstruction methods with different representations under the public datasets were compared and analyzed.Finally,the difficulties and challenges introduced by single image-based 3d reconstruction were discussed,and some future researches were also proposed.