2020/05/11 by Junsoo Lee, Eungyeup Kim, Lee, Junsoo +9 · 3 citations
Computer Science · Economics, Econometrics and Finance · #Cinema and Media Studies #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Generative Adversarial Networks and Image Synthesis
paper · pdf · doi:10.48550/arxiv.2005.05207
openalex publication_date 2020/05/11 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
This paper tackles the automatic colorization task of a sketch image given an\nalready-colored reference image. Colorizing a sketch image is in high demand in\ncomics, animation, and other content creation applications, but it suffers from\ninformation scarcity of a sketch image. To address this, a reference image can\nrender the colorization process in a reliable and user-driven manner. However,\nit is difficult to prepare for a training data set that has a sufficient amount\nof semantically meaningful pairs of images as well as the ground truth for a\ncolored image reflecting a given reference (e.g., coloring a sketch of an\noriginally blue car given a reference green car). To tackle this challenge, we\npropose to utilize the identical image with geometric distortion as a virtual\nreference, which makes it possible to secure the ground truth for a colored\noutput image. Furthermore, it naturally provides the ground truth for dense\nsemantic correspondence, which we utilize in our internal attention mechanism\nfor color transfer from reference to sketch input. We demonstrate the\neffectiveness of our approach in various types of sketch image colorization via\nquantitative as well as qualitative evaluation against existing methods.\n