Reviews: Soft-Gated Warping-GAN for Pose-Guided Person Image Synthesis

This paper presents a parsing-guided deep network for re-posing a person image. In particular, the network takes as inputs a person image and a novel pose and then produces a novel image of the same person in the novel pose. The proposed network consists of two stages. In the first stage, a off-the-shelf human parser is used to segment the input image into parts. Given the part segmentation and the target pose, a pix2pix network is trained to synthesize target part segmentation masks or parsing.