Comicolorization : Semi-automatic Manga Colorization

Given a monochrome manga and reference images as inputs, a colorization tool, "Comicolorization", generates a plausible color version of the manga. Although colorization research has a long history, no approach has focused on the colorization of an entire manga. We present three contributions. First, we propose the first system to colorize an entire manga at once. Our method colorizes a whole page (not a single panel). Second, our tool can choose vivid colors while maintaining the color coherency between multiple panels. To achieve coherency and vividness, we input the color features to help the inference by our deep colorization network. The color features are extracted from inputted reference images. Our approach employs adversarial loss while training the colorization network to encourage the effect of the color features. Finally, our tool allows users to revise the colorization result interactively. They can control color bleeding through adjusting color features manually. Moreover, they can put the color dots as hints for the desired colorization.