Auto-Adaptivity: An Optimization-Based Approach to Spatial Adaptivity for Smoke Simulations

Abstract

Figure 1: Our new approach to spatial adaptivity enables the user to run adaptive simulations of smoke that are visually close to identical to their sparse non-adaptive counterpart (a) with the benefit of a reduction in computation-time and memory. A few input parameters (b) are fed into our new auto-adaptivity algorithm that retains the voxels which — subject to the constraint of a user-specified computation-budget (fidelity) — globally maximizes the quality of the simulation according to criteria such as distortion-rate, detail and interpolation error. The auto-adaptivity algorithm frees the user from explicitly managing and combining adaptivity controls by automatically determining which voxels to coarsen and refine as shown in the dirt bike simulation which represents the dust at four different levels of resolution (c). Dirt bike courtesy of Kerosene VFX.