Pose- and Attribute-consistent Person Image Synthesis

Person Image Synthesis aims at transferring the appearance of the source person image into a target pose. Existing methods cannot handle large pose variations and therefore suffer from two critical problems: (1) synthesis distortion due to the entanglement of pose and appearance information among different body components and (2) failure in preserving original semantics (e.g., the same outfit). In this article, we explicitly address these two problems by proposing a Pose- and Attribute-consistent Person Image Synthesis Network (PAC-GAN). To reduce pose and appearance matching ambiguity, we propose a component-wise transferring model consisting of two stages. The former stage focuses only on synthesizing target poses, while the latter renders target appearances by explicitly transferring the appearance information from the source image to the target image in a component-wise manner. In this way, source-target matching ambiguity is eliminated due to the component-wise disentanglement of pose and appearance synthesis. Second, to maintain attribute consistency, we represent the input image as an attribute vector and impose a high-level semantic constraint using this vector to regularize the target synthesis. Extensive experimental results on the DeepFashion dataset demonstrate the superiority of our method over the state of the art, especially for maintaining pose and attribute consistencies under large pose variations.

Automated summary

Person Image Synthesis addresses two critical problems in transferring appearance of a source person image to a target pose: synthesis distortion due to pose and appearance entanglement, and failure in preserving original semantics. The proposed PAC-GAN explicitly tackles these problems by using a component-wise transferring model and a high-level semantic constraint. Experimental results on DeepFashion dataset demonstrate the superiority of our method in maintaining pose and attribute consistencies.