Can adversarial networks hallucinate occluded people with a plausible aspect?

作者：

Highlights：

• We propose a novel Adversarial Network that solves occlusions in pedestrian images.

• Our reconstruction keeps both the appearance and the background coherent.

• We condition the occluded body part restoration on pedestrian attributes.

• We devise a new way for synthetically generating occlusion and visible pairs.

• We provide a huge CG dataset for pedestrian attribute recognition in crowded areas.

摘要

•We propose a novel Adversarial Network that solves occlusions in pedestrian images.•Our reconstruction keeps both the appearance and the background coherent.•We condition the occluded body part restoration on pedestrian attributes.•We devise a new way for synthetically generating occlusion and visible pairs.•We provide a huge CG dataset for pedestrian attribute recognition in crowded areas.

论文关键词：GAN,Attribute recognition,Occlusions

论文评审过程：Received 3 July 2018, Revised 22 January 2019, Accepted 23 March 2019, Available online 27 March 2019, Version of Record 17 April 2019.

论文官网地址：https://doi.org/10.1016/j.cviu.2019.03.007