Face-parsing-constrained CycleGAN for Personalized Facial Paper-cut Generation
Yonghai Du, Jingzhou Huang, Taixiang Zhang, Houpan Zhou · 2024
Facial paper-cutting, a cherished aspect of Chinese folk art, is facing the risk of diminishing due to the complexities of manual creation and the time-consuming production process. This paper proposes a novel approach for generating facial paper-cuts by employing a CycleGAN constrained by face parsing. Given the fundamental attributes of facial paper-cuts—minimalist design and intentional feature exaggeration—direct application of CycleGAN may lead to inaccuracies in facial features and unintended shadow effects. An attention mechanism, integrating facial semantic segmentation with channel attention, is introduced to refine the generator’s output. Comparative evaluations reveal that this method produces facial paper-cuts of superior quality compared to existing techniques. This novel approach has the potential to preserve and promote the unique art of Chinese papercut in the digital age.