Reconstructing Production Data from Traditional Japanese Animation (DAAV)

The creation of traditional animation is performed in multiple steps, creating various intermediary products. Image processing and machine learning could be used for the automation of some of these time-consuming steps to help animators and studios. However, machine learning solutions require large amounts of example data, which is not available for the intermediary products of animation. On the other hand, final animation is more easily available through public datasets, video releases and streaming services. This work aims to bridge this gap by creating a tool to predict intermediary products of animation from frames of the final video, using machine learning. The predicted production data can then be used for further research.

In particular, frames of traditional animation are made out of background and foreground elements, which are produced through different workflows. Foreground elements are created by first creating color coded lineart. It will be attempted to automatically segment frames into background and foreground. The predicted foreground will then be used to attempt the automatic recreation of the corresponding lineart.