Stereoscopic Clustered Light Shading

Abstract

Real-time realistic rendering requires the evaluation of the influence of many light sources. In case of dynamic geometry or light sources, this evaluation must be performed every frame. In this thesis I present enclosed clustering: an adaptation of Clustered Light Shading to stereoscopic rendering which cuts the per-eye cost of light assignment in half. To achieve this, clustering is performed once with a clustering camera frustum that encloses both of the eye camera frusta. Decoupling of the clustering camera from the rendering camera gives way to two alternative ways of constructing the clustering: orthographic clustering and displaced perspective clustering. Orthographic clustering uses a uniform world-space grid and has been traditionally dismissed. Displaced perspective clustering uses a decoupled clustering camera to reduce the number of small clusters near the camera. These techniques can be applied to both monoscopic and stereoscopic rendering.