UC Berkeley

This paper describes a method for view-dependent cloth simulation using dynamically adaptive mesh refinement and coarsening. Given a prescribed camera motion, the method adjusts the criteria controlling refinement to account for visibility and apparent size in the camera's view. Objectionable dynamic artifacts are avoided by anticipative refinement and smoothed coarsening. This approach preserves the appearance of detailed cloth throughout the animation while avoiding the wasted effort of simulating details that would not be discernible to the viewer. The computational savings realized by this method increase as scene complexity grows, producing a 2× speed-up for a single character and more than 4× for a small group.