Lighting-driven voxels for memory-efficient computation of indirect illumination

摘要

Several recently proposed voxel-based global illumination algorithms rely on the use of reflective shadow maps (RSMs) to interactively compute indirect illumination. However, RSMs do not scale well with the number of light sources because of their high memory consumption when rendering. Observing that, in most cases only a fraction of the voxels really contribute to single-bounce indirect illumination, in this paper we propose the use of lighting-driven voxels (LDVs), which are constructed from a subset of voxels, to reduce the memory burden. They are used in conjunction with a voxel-based global illumination algorithm that enables the interactive indirect illumination of dynamic scenes. We evaluate the memory usage, query performance, and construction speed for various voxel resolutions. Empirically, rendering with LDVs consumes an order of magnitude less memory than rendering with RSMs. Further, it achieves a higher performance for radiance queries when multiple light sources are used. Moreover, we integrated our method into voxel ray tracing and voxel cone tracing. For each of algorithm, we achieve an interactive performance that significantly reduces memory with respect to the reference solution.