One-to-many

We propose an example-based approach for synthesizing diverse mesh animations. Provided a short clip of deformable mesh animation, our method synthesizes a large number of different animations of arbitrary length. Combining an automatically inferred linear blending skinning (LBS) model with a PCA-based model reduction, our method identifies possible smooth transitions in the example sequence. To create smooth transitions, we synthesize reduced deformation parameters based on a set of characteristic key vertices on the mesh. Furthermore, by analyzing cut nodes on a graph built upon the LBS model, we are able to decompose the mesh into independent components. Motions of these components are synthesized individually and assembled together. Our method has the complexity independent from mesh resolutions, enabling efficient generation of arbitrarily long animations without tedious parameter tuning and heavy computation. We evaluate our method on various animation examples, and demonstrate that numerous diverse animations can be generated from each single example.