“Controllable motion textures” by Wang, Hsiao, Chen and Huang

While producing an animation, we often rely on the artist to make it manually, or sometimes we generate the animation by some physical simulations. However, the animation we created by these methods always has fixed length and is very difficult to reuse. For 3D animation production, in practice, we always need some animations with variable lengths, for example, we would like to place some flags waving in the wind as part of the background, also perhaps nearby some trees standing with their swaying branches. It will be significantly time-consuming and inefficient to make fixed-length animations (flags and trees) for a specified scene every time when the scene is altered. Consequently, suppose we have produced a segment of an animation, then based on some analysis of it, we could randomly generating a new animation, which length is specified as user’s pleases. For example, we could generate an animation which shows a square surrounding by tens of flags swinging differently by using only one short flag animation originally. By doing this, the time for making films could considerably be saved. In this paper, we transform the input animation into a medium called motion texture, which could be used to generate random-length animations. A motion texture records lots of parts of an animation and the relations between these segments. According to the information, we could generate a large amount of different animations we require in the future.

1. Schödl, A., Szeliski, R., Salesin, D. H., and essa, I. 2000. Video textures. In SIGGRAPH 2000, 489–498.