Expediting Precomputation for Reduced Deformable Simulation

摘要

Model reduction has popularized itself for simulating elastic deformationfor graphics applications. While these techniques enjoyorders-of-magnitude speedups at runtime simulation, the efficiencyof precomputing reduced subspaces remains largely overlooked.We present a complete system of precomputation pipelineas a faster alternative to the classic linear and nonlinear modalanalysis. We identify three bottlenecks in the traditional modelreduction precomputation, namely modal matrix construction,cubature training, and training dataset generation, and accelerateeach of them. Even with complex deformable models, our methodhas achieved orders-of-magnitude speedups over the traditionalprecomputation steps, while retaining comparable runtime simulationquality.