Style-Content-Aware Adaptive Normalization Based Pose Guided for Person Image Synthesis
Wei Wei, Xia Yang, Xiaodong Duan, Chen Guang Guo · IEEE Access · 2023
Most of the tasks based on pose-guided person image synthesis have obtained accurate target pose, but still have not obtained reasonable style texture mapping. In this paper, we propose a new two-stage network to decouple style and content, which aims to enhance the accuracy of pose transfer and the realism of a person appearance. Firstly, we propose an Aligned Multi-scale Content Transfer Network(AMSNet) to predict the target edge map for pose content transfer in advance, which can not only preserve clearer texture content but also alleviate spatial misalignment through advancing to transfer pose information. Secondly, we propose a new Style Texture Transfer Network(STNet) to transfer gradually the source style features to the target pose to achieve a reasonable distribution of styles. In order to make the appearance texture highly similar to the source style, a style-content-aware adaptive normalization method was proposed. The source style features were mapped into the same latent space with aligned content images(the target pose and edge), and the consistency of the style texture and content was enhanced through adaptive adjustment of the source style and target pose. To enhance the realism of the target generated texture through preserving the source style features. The experimental results show, both quantitatively and qualitatively, that the model can synthesize target images that are consistent with the source style and obtain superior results.