Image synthesis via adversarial geometric consistency pursuit

作者：

Highlights：

• Geometric distortion and appearance difference are difficult for image synthesis.

• Warping images in a learned transformation improves the synthesis result.

• A robust network with a geometric consistency pursuit strategy is effective.

• Appearance harmonization achieves seamless image synthesis.

摘要

•Geometric distortion and appearance difference are difficult for image synthesis.•Warping images in a learned transformation improves the synthesis result.•A robust network with a geometric consistency pursuit strategy is effective.•Appearance harmonization achieves seamless image synthesis.

论文关键词：Image synthesis,Geometric consistency,Appearance harmonization,Generative adversarial network,Spatial transformer

论文评审过程：Received 31 January 2020, Revised 29 August 2021, Accepted 10 September 2021, Available online 17 September 2021, Version of Record 24 September 2021.

论文官网地址：https://doi.org/10.1016/j.image.2021.116489