Deep Learning Applications for the Production of Animation

• In Event: Faculty Research-in-Progress Digital (Poster) Presentations - Session V

Abstract

Using Deep Learning algorithms to explore and develop new ways of doing animation is currently both a technical and aesthetic challenge. In this paper I present one approach of incorporating machine learning into the animator’s toolset. Animation production has often been a leader in the adoption of technology in the motion picture industry, so exploring the emerging technology of machine and deep learning seems a good fit. Machine learning, defined by the IEEE as “the study of computer algorithms that improve automatically through experience,” opens up the possibilities for computing devices to become adept at skills difficult or impossible to specify in traditional computer code. Its refinement, “Deep learning” is a form of machine learning that enables computers to learn from data sets and understand them in terms of a conceptual hierarchy. Deep Learning algorithms have shown capacity for creating and modifying the still image (for example, Google’s Deep Dream algorithm which introduces one image’s ”style” into another image’s content). However less work has been done in the direction of the moving image, in particular, animation. My research shows that Deep Learning networks can provide rapid generation of animated backgrounds under the supervision of human animators.

Much as the addition of computer graphics tools ignited a renaissance of animation in the past twenty years, the development of Deep Learning tools may spark additional flourishing in the art form. Deep Learning provides an unprecedented approach to the creation of computer animation, which may yield groundbreaking results for story-driven computer graphics.

Author

• John Perkins-Buzo, Southern Illinois University @ Carbondale