Three-view generation based on a single front view image for car
- PDF / 3,168,660 Bytes
- 11 Pages / 595.276 x 790.866 pts Page_size
- 14 Downloads / 177 Views
ORIGINAL ARTICLE
Three-view generation based on a single front view image for car Zixuan Qin1 · Mengxiao Yin1,2 · Zhenfeng Lin1 · Feng Yang1,2 · Cheng Zhong1,2 Accepted: 8 September 2020 © Springer-Verlag GmbH Germany, part of Springer Nature 2020
Abstract The multi-view of an object can be used for 3D reconstruction. The method proposed in this paper generates the left and the top view of a target car through deep learning. The input of the method is only a front view of a 3D car and it isn’t necessary for the depth of the 3D car. Firstly, a rough orthographic views of the 3D car is gotten from an information constraint network which is constructed by considering the structural relation between one view and the other two views. And then the rough orthographic views is transformed into large-pixel block rough view through the nearest interpolation, at the same time, the large-pixel blocks are also migrated to improve the quality of the rough orthographic views. Finally, the generative adversarial network with perception loss is used to enhance the large-pixel block view. In addition, the three views generated by the network can be used to synthesize a 3D point cloud shell. Keywords Deep learning · Orthographic views · Front view · Point cloud shell
1 Introduction Three-view drawing (i.e. orthogonal projection method) has been used widely for a long time, which plays an important role in engineering design [8,20]. And it is a common method for describing 3D objects with 2D information in engineering drawings. For image processing, the three-view images of an object can be obtained by photographing from the front, the left and the top of the object. Furthermore, Tatarchenko et al. [35] called all views of a 3D object and corresponding depth maps as “multi-view 3D model”. It is more difficult to generate the left view, the top view and even multi views image from a single front view in the traditional technology, since the corresponding 3D model information is very vague when there is only a front view image. Currently, the difficulty of this task can be greatly reduced by deep learning. Deep learning has become a tool which is widely used in many fields. Unsupervised or weakly supervised learning is applied to reduce the dependence of the segmentation method on data labeling and improve segmentation results [19,29,31]. Shu et al. [30] employed a deep
B
Mengxiao Yin [email protected]
1
School of Computer and Electronic Information, Guangxi University, Nanning, China
2
Guangxi Key Laboratory of Multimedia Communications and Network Technology, Nanning, China
learning method based on multiple feature descriptors to extract 3D points of interest. Luciano et al. [22] proposed a scalable fusion framework for graph-structured data at the foundation of graph convolutional neural network. Deeper feature representations of 3D shapes may be learned and given by this network. Zhang et al. [39] designed a network to generate high-quality feature vector and 3D scenes of different styles can be produced by the network. To impr
Data Loading...
