Alec Rivers MIT CSAIL Takeo Igarashi The University of Tokyo Fr´ do Durand e MIT CSAIL
Figure 1: A 2.5D Cartoon: We take vector art drawings of a cartoon from different views (a) and use them to automatically generate a 2.5D cartoon (b), which associates each stroke with a 3D position. The 2.5D cartoon can then be used to simulate a rotation in 3D andgenerate a rendering of the cartoon in a novel view (c).
We present a way to bring cartoon objects and characters into the third dimension, by giving them the ability to rotate and be viewed from any angle. We show how 2D vector art drawings of a cartoon from different views can be used to generate a novel structure, the 2.5D cartoon model, which can be used to simulate 3D rotationsand generate plausible renderings of the cartoon from any view. 2.5D cartoon models are easier to create than a full 3D model, and retain the 2D nature of hand-drawn vector art, supporting a wide range of stylizations that need not correspond to any real 3D shape. Keywords: non-photorealistic rendering, cartoons, vector art, billboards, animation, interpolation
as the 2D appearance intended maynot correspond to any real 3D shape. We propose a way to use 2D vector art drawings of a cartoon from different angles to generate a new type of structure, the 2.5D cartoon model, visualized in Figure 1 (b). This structure associates each stroke of a cartoon with a single 3D position, and is able to generate plausible renderings of the cartoon in new views by translating the strokes’ positions in3D, while interpolating their shapes in 2D. We have found that this simple model structure allows surprisingly believable rotations of cartoons, despite having no explicit 3D polygonal mesh of the object, and typically only requires three or four deﬁned views of the cartoon to be able to generate plausible renderings of the cartoon in any orientation. The result is a cartoon that retains the 2D,hand-drawn nature of the input vector art, while supporting full 3D rotation.
The ability to rotate a cartoon and view it from any angle has wideranging applications, both in aiding animation and in enabling cartoon objects to be placed in an interactive 3D environment in which the user controls the viewpoint. Previously, this has been achieved by constructing a 3D model of thecartoon and rendering it in a non-photorealistic way so as to resemble a cartoon. However, generating a 3D model is time-consuming, and many stylistic elements of 2D drawings cannot be adequately reproduced in a 3D model,
2 Related Work
Cartoon-style drawing and animation have over time developed a rich language of non-photorealistic stylizations (see, e.g., [Johnston and Thomas 1981; Blair1994]). Much work has been done that seeks to introduce these stylizations to computer rendering (see [Gooch and Gooch 2001] for a survey). Non-photorealistic rendering techniques have been proposed to make a 3D model resemble a hand-drawn cartoon, such as cel-shaded lighting [Decaudin 1996] or exaggerated, 2D silhouettes [Northrup and Markosian 2000; Kalnins et al. 2002]. However, rendering from asingle 3D model cannot account for cartoons that have mutually inconsistent appearances in different views – for example, no matter the viewing angle, Bugs Bunny’s ears are always facing the camera. View-dependent geometry [Rademacher 1999] addresses this limitation by providing multiple different models of an object, each associated with a single perspective, and rendering intermediateperspectives using a 3D model that is an interpolation of the models for nearby orientations. This approach, while more ﬂexible, requires additional 3D modeling, and retains the essential 3D nature and appearance of the model.Our approach,
Figure 2: 2.5D interpolation: Here we show a 2.5D model being rotated through (a) front, (b) oblique, and (c) side views. The arrangement of...