Abstract
Recently, several soft shadow mapping algorithms have been introduced which extract micro-occluders from a shadow map and backproject them on the light source to approximately determine light visibility. To maintain real-time frame rates, these algorithms often have to resort to coarser levels of a multi-resolution shadow map representation which can lead to visible quality degradations. In particular, discontinuity artifacts can appear when having to use different shadow map levels across pixels. In this paper, we discuss several aspects of soft shadow quality. First, we motivate and propose a scheme that allows for varying soft shadow quality in screen-space in a visually smooth way and also for its adaptation based on local features like assigned importance. Second, we suggest a generalization of micropatches which yields a better occluder geometry approximation at coarser shadow map levels, thus helping to reduce occluder overestimation. Third, we introduce a new hybrid acceleration structure for pruning the search space of potential micro-occluders that enables employing finer shadow map levels and hence increasing quality. Finally, we address multisampled rendering and suggest a simple scheme for interpolating light visibility that only adds a negligible cost compared to single-sample rendering.
