Expediting Precomputation for Reduced Deformable Simulation

Yin Yang, Dingzeyu Li, Weiwei Xu, Yuan Tian, Changxi Zheng

ACM Transactions on Graphics (SIGGRAPH Asia 2015), 34(6)

Model reduction has popularized itself for simulating elastic deformation for graphics applications. While these techniques enjoy orders-of-magnitude speedups at runtime simulation, the efficiency of precomputing reduced subspaces remains largely overlooked. We present a complete system of precomputation pipeline as a faster alternative to the classic linear and nonlinear modal analysis. We identify three bottlenecks in the traditional model reduction precomputation, namely modal matrix construction, cubature training, and training dataset generation, and accelerate each of them. Even with complex deformable models, our method has achieved orders-of-magnitude speedups over the traditional precomputation steps, while retaining comparable runtime simulation quality.

Paper / Paper (low resolution)
Video (145MB) / Youtube

bibtex citation
  title={Expediting Precomputation for Reduced Deformable Simulation},
  author={Yang, Yin and Li, Dingzeyu and Xu, Weiwei and Tian, Yuan and Zheng, Changxi},
  journal={ACM Trans. Graph.},

We thank anonymous reviewers for their feedback. This research was supported in part by the National Science Foundation (CRII-1464306, CAREER-1453101), National Science Foundation China (No. 61272392, No. 61322204), UNM RAC & OVPR research grants, and generous gifts from Intel and DJI. Any opinions, findings and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of funding agencies or others.