Adaptive Multi-level Multi-phenomena Appearance Models

点击翻译按钮获取中文摘要

Rendering realistic synthetic scenes requires an accurate representation of 3D objects and light transport. The result of this simulation is filtered in an image as surface appearance, which varies with distance, orientation, incident illumination, reflection properties, interreflections, subsurface scattering, microgeometry, and textures. Light transport and visibility are often resolved using a combination of Monte Carlo (MC) sampling and ray tracing, well adapted through multiple sampling to most integration problems in rendering. However, they are inefficient in complex scenes because they operate on the finest-scale of the scene representation, and thus, this approach does not scale for real-time rendering of complex scenes nor offline high-quality rendering. Instead, we propose to adapt pre-filtered representations of geometries, visibility, and appearances to the pixel size, i.e., to find an appropriate level-of-detail computation for light-object interactions. Our recent LEADR method computes the color over an entire pixel footprint, interactively and independent of the footprint's size. It is filterable, efficient, suitable for GPUs, scalable with complexity, and high-quality, but it is limited to textured height maps under direct illumination. This research program will design several novel hierarchical representations of surfaces, volumes, visibility/transparency, textures, and light transport, to cast the phenomena that affect appearance into our prefiltered approach. We propose to generalize the concept of light transport within a black box to all combined factors affecting appearance, thus efficiently determining the appropriate representation under viewing and illumination conditions, leading to real-time rendering of complex scenes, as well as an adapted scheme to be integrated in path-tracing methods. We will build knowledge from a number of specific case studies, such as grass-to-lawn, tree-to-canopy, sand-to-dune, fabrics-to-drapes, building-to-city, etc. We will develop interpolation schemes between two adjacent levels in a hierarchy. Then we will work our way up by introducing representations at coarser hierarchy levels, and adapting interpolation schemes. Next we will add the other effects affecting appearance, integrating them into a more general model. Our long-term goal is to develop a unified multi-level multi-phenomena appearance scheme suitable for high-quality real-time and offline rendering.