Line Drawings for Face Portraits From Photos Using Global and Local Structure Based GANs

Despite significant effort and notable success of neural style transfer, it remains challenging for highly abstract styles, in particular line drawings. In this paper, we propose APDrawingGAN++, a generative adversarial network (GAN) transforming face photos to artistic portrait drawings (APDrawings), which addresses substantial challenges including style, different drawing techniques facial features, high perceptual sensitivity artifacts. To address these, composite GAN architecture that consists local networks (to learn effective representations specific features) global capture the overall content). We provide theoretical explanation necessity structure by proving any with single generator cannot generate styles like APDrawings. further introduce classification-and-synthesis approach lips hair where are used artists, applies suitable given input. art form inherent APDrawings, two operations-(1) coping lines small misalignments while penalizing large discrepancy (2) generating more continuous lines-by introducing novel loss terms: one is distance transform nonlinear mapping other continuity loss, both improve quality. also develop dedicated data augmentation pre-training results. Extensive experiments, user study, show our method outperforms state-of-the-art methods, qualitatively quantitatively.