Human Pose Transfer with Depth Information

Human pose transfer refers to synthesizing an image of a target person with the pose of a source person. Many current methods have achieved good results. However, the images generated on a two-dimensional plane will cause confusion in the positions of different body parts due to the lack of depth information in three dimensions, which limits the visual quality of the results. This paper introduces monocular depth estimation to solve this problem and proposes a new framework that can predict the depth information of the human body and thus synthesize high-quality target human appearance.