EdgeFont: Enhancing style and content representations in few-shot font generation with multi-scale edge self-supervision

Yefei Wang, Kangyue Xiong, Yiyang Yuan, Jinshan Zeng

Abstract: Highlights•A novel few-shot font generation model is proposed by integrating edge information.•Edge information is used by a self-supervised way without manual edge labeling.•The self-supervised module can enhance the style and content representations.•Effectiveness of proposed model is demonstrated by extensive experiments.•Good generalization of proposed model for cross-lingual font generation is shown.