A Novel View Image Generation Network Based On Attention Mechanism Refining Features

We propose a novel view image generation network based on attention mechanism refining features. According to the prior structural model learned in the training, the network can generate multi-view images from a single-view image of an object we are interested in. The input of our network is the RGB...

Full description

Saved in:
Bibliographic Details
Published in2021 International Conference on Control, Automation and Information Sciences (ICCAIS) pp. 968 - 973
Main Authors Yi, Mengni, Hui, Bingwei, Hu, Weidong, He, Min
Format Conference Proceeding
LanguageEnglish
Published IEEE 14.10.2021
Subjects
adversarial learning
attention mechanism
Face recognition
Image synthesis
novel view generation
Refining
skip connection
Target recognition
Technological innovation
Three-dimensional displays
Training
Online AccessGet full text

Cover

More Information
Summary:We propose a novel view image generation network based on attention mechanism refining features. According to the prior structural model learned in the training, the network can generate multi-view images from a single-view image of an object we are interested in. The input of our network is the RGB image and pose corresponding to the source viewpoint, as well as the target pose we desired, and the output is the predicted target view image. The main innovation of our method is to exploit the attention mechanism to help us select better and more important features by assigning weight to features. Our method needs no additional 3D supervision and balances the accuracy and speed well. The final quantitative and qualitative results show that our network has achieved impressive results in the task of novel view image generation.
ISSN:2475-7896
DOI:10.1109/ICCAIS52680.2021.9624598