Expediting Precomputation for Reduced Deformable Simulation

摘要

Model reduction has popularized itself for simulating elastic deformationrnfor graphics applications. While these techniques enjoyrnorders-of-magnitude speedups at runtime simulation, the efficiencyrnof precomputing reduced subspaces remains largely overlooked.rnWe present a complete system of precomputation pipelinernas a faster alternative to the classic linear and nonlinear modalrnanalysis. We identify three bottlenecks in the traditional modelrnreduction precomputation, namely modal matrix construction,rncubature training, and training dataset generation, and acceleraterneach of them. Even with complex deformable models, our methodrnhas achieved orders-of-magnitude speedups over the traditionalrnprecomputation steps, while retaining comparable runtime simulationrnquality.