Yun Kwan ; Seo Kwanggyoon ; Seo Chang Wook ; Yoon Soyeon ; Kim Seongcheol ; Ji Soohyun ; Ashtari Amirsaman ; Noh Junyong

Abstract

Facial sketches are both a concise way of showing the identity of a personand a means to express artistic intention. While a few techniques have recentlyemerged that allow sketches to be extracted in different styles, they typicallyrely on a large amount of data that is difficult to obtain. Here, we proposeStyleSketch, a method for extracting high-resolution stylized sketches from aface image. Using the rich semantics of the deep features from a pretrainedStyleGAN, we are able to train a sketch generator with 16 pairs of face and thecorresponding sketch images. The sketch generator utilizes part-based losseswith two-stage learning for fast convergence during training for high-qualitysketch extraction. Through a set of comparisons, we show that StyleSketchoutperforms existing state-of-the-art sketch extraction methods and few-shotimage adaptation methods for the task of extracting high-resolution abstractface sketches. We further demonstrate the versatility of StyleSketch byextending its use to other domains and explore the possibility of semanticediting. The project page can be found inhttps://kwanyun.github.io/stylesketch_project.