Advancing Facial Stylization through Semantic Preservation Constraint and Pseudo-Paired Supervision

TL;DR

This paper introduces a novel facial stylization method that employs semantic preservation constraints and pseudo-paired supervision to improve content fidelity and stylization quality, outperforming previous approaches.

Contribution

It proposes a new framework combining semantic preservation and pseudo-paired datasets for enhanced facial stylization, enabling flexible multimodal and reference-guided stylization.

Findings

Produces high-fidelity stylized portraits with fewer artifacts.

Outperforms previous methods in aesthetic quality and content preservation.

Enables flexible multimodal and reference-guided stylization without complex architectures.

Abstract

Facial stylization aims to transform facial images into appealing, high-quality stylized portraits, with the critical challenge of accurately learning the target style while maintaining content consistency with the original image. Although previous StyleGAN-based methods have made significant advancements, the generated results still suffer from artifacts or insufficient fidelity to the source image. We argue that these issues stem from neglecting semantic shift of the generator during stylization. Therefore, we propose a facial stylization method that integrates semantic preservation constraint and pseudo-paired supervision to enhance the content correspondence and improve the stylization effect. Additionally, we develop a methodology for creating multi-level pseudo-paired datasets to implement supervisory constraint. Furthermore, building upon our facial stylization framework, we achieve more flexible multimodal and reference-guided stylization without complex network architecture designs or additional training. Experimental results demonstrate that our approach produces high-fidelity, aesthetically pleasing facial style transfer that surpasses previous methods.