Real-time rendering of continuous levels of detail for sparse voxel octrees

Abstract

In this paper, we present a novel approach to real-time, continuous and symmetrical level of detail (LOD) management of a 3D object represented by a sparse voxel octree (SVO). We propose a new method for continuous and symmetrical transition between two detail levels. The method is based on a SVO representation extended by redundant, helper nodes which are used to achieve a proper interpolation of geometry and material data. We extend redundant nodes with a transition direction attribute. Additional memory requirements are minimized by storing indices in a direction vector lookup table in object space. The new method is applied for an accurate evaluation of the required LOD. It uses an image-based evaluation function, i.e. the standard level transition function based on camera distance is extended by the real-time calculation of the current LOD pixel fill rate. We extend typical level transition function based on distance with real-time calculation of the current LOD pixel fill rate. Two different image based methods of SVO node pixel fill rate calculations using compute shaders or GPU queries and parallel reduce are presented. The developed LOD management algorithm is applicable for a raytracing and a rasterizationbased rendering pipeline. The LOD transition algorithm allows to perform a dynamic and continues control of the SVO based objects which have not been available in other works. Moreover, the proposed fading algorithm based on the fade out direction and scaling allows for a LOD change without any graphical artifacts or loss of the virtual scene immersion.