Coherent Out‐of‐Core Point‐Based Global Illumination

Abstract

Figure 1: A cityscape from the DreamWorks Animation movie "Kung Fu Panda 2". The out-of-core method described in our paper shaded the global illumination for the whole frame in 6 minutes 23 seconds. It used 128 million points, and it took an additional 4 minutes 18 seconds to build the out-of-core octree, resulting in 27 million octree nodes. The total amount of outof-core data was 6.8 GB while the in-core point and node caches' memory usage did not exceed 2 GB. The cache hit rate was 99%. In Section 7, we demonstrate usage of up to 1.7 billion points (88 GB of data) within a 2 GB memory cap. AbstractWe describe a new technique for coherent out-of-core point-based global illumination and ambient occlusion. Point-based global illumination (PBGI) is used in production to render tremendously complex scenes, so in-core storage of point and octree data structures quickly becomes a problem. However, a simple out-of-core extension of a classical top-down octree building algorithm would be extremely inefficient due to large amount of I/O required. Our method extends previous PBGI algorithms with an out-of-core technique that uses minimal I/O and stores data on disk compactly and in coherent chunks for later access during shading. Using properties of a space-filling Z-curve, we are able to preprocess the data in two passes: an external 1D-sort and an octree construction pass.

GPU-based out-of-core many-lights rendering

“…Surface samples are generated by an out-of-core ray tracer that traces rays from supersamples of pixels. There are several ways to generate VPLs fast and efficiently, such as a standard VPLs generation [Keller 1997] using an out-of-core ray tracer or sampling mesh surfaces by randomly distributing samples in triangles [Kontkanen et al 2011]. We generated surface samples and VPLs in a preprocess and fed them with geometry as input for our algorithm.…”

“…Our main strength is the scalability and efficiency for rendering global illumination effects on massive models with a large number of point lights. It has been demonstrated that hundreds of million or billion VPLs are needed for movie-quality rendering [Kontkanen et al 2011]. Moreover, with the rapid increase of captured geometry data size, the data size of lights need to be increased to capture detailed illumination for small-scale geometric features.…”

“…The point-based representation of lights also appeals to the film industry. The point-based global illumination method (PBGI) [Christensen 2008] and its out-of-core version [Kontkanen et al 2011] have been used in film production. Besides these offline rendering approaches, Ritschel et al [2009] computed global illumination by rasterizing point-based lights on thousands of tiny micro-buffers and achieved interactive rendering rates.…”

“…Such a motivation bears some similarity with the out-of-core PBGI [Kontkanen et al 2011]. However, the surfels approximation and inaccurate visibility tests used in PBGI ease the challenges in data management and have less flexibility to tradeoff between quality and performance.…”

Reconstructing the indirect light field for global illumination

“…To accelerate this, we organize the secondary hit points {hi} into a motion bounding volume hierarchy [Glassner 1988] using the streaming algorithm of Kontkanen et al [2011]. During reconstruction, the hierarchy is used in a standard fashion by walking the BVH and intersecting the reconstruction ray with the splats in the intersected leaf nodes.…”