endingcredits Simulate. Everything.

Abstract

Cartoon shading can be a long and tedious task when creating an animated work of even minute length. In this paper, we attempt to automate the process by having an artist shade a 2-D image of a small portion or limited view of an object. We then attempt to learn the stylistic elements, or characteristics of the artist, using supervised learning algorithms and then apply that knowledge to shade in 2-D the rest of the object, the object from another view, and / or other objects altogether. We present renderings of a variety of 3D models, produced using learned models of artist style. We show that neural networks are a effective means of learning such styles and provide a comparative analysis against another supervised learning algorithm, k-Nearest Neighbors