Semi-Automatic Colorization Pipeline for Anime Characters and its Evaluation in Production

Abstract

Improving the efficiency of a colorization process in anime productions is necessary to enhance the quality of animes. In this paper, we introduce a semi-automatic anime character colorization pipeline based on few-shot patch-based learning which is specific to the anime style. Our pipeline requires only a small number of line-drawings and their colorized images, which is intermediate products in their current workflow as training data. The advantage of our method is that it is possible to complete the training process of a sequence-specific colorization model on the fly. To evaluate the effectiveness of our pipeline, we conduct a questionnaire survey for colorization artists after several trials of our colorization pipeline in an actual anime production. As a result, our pipeline has proven to be effective in improving the colorization process efficiency.