Efficient Screenspace Rendering for Area Lights
Koa, Ming Di
Date of Issue2016
Proceedings of the 33rd Computer Graphics International (CGI '16)
School of Computer Science and Engineering
Fraunhofer IDM Centre@NTU
Efficient rendering of illumination from area lights in virtual scenes has always proved to be challenging. We extend the work of multi resolution rendering and Light Propagation Volumes (LPV) to simulate direct and indirect illumination from area lights respectively. To compute direct illumination, we create 2D multi resolution fragments to represent the scene on the screenspace, in which higher resolution fragments are created when normal, depth and visibility discontinuity are found. Our subdivision scheme performs a sub-fragment visibility test (SFVT) within each fragment and our proposed gradient aware screenspace subdivision (GASS) algorithm accelerates the refinement by increasing the number of subdivisions based on gradient differences. We also propose a single pass screenspace irradiance up-sampling scheme which uses Gaussian radial basis functions (RBF) for interpolating scattered fragments. This reduces artifacts caused by large fragments while also significantly reducing the number of fragments that we require. Our indirect illumination is computed by distributing a set of Poisson sample points in the scene. Each LPV voxel performs a light gathering operation on these samples and deposits them internally. Light intensity in the LPV is propagated simulating indirect illumination from area lights. From experiments, our techniques are able to run at interactive rates.
© 2016 The author(s), published by ACM. This is the author created version of a work that has been peer reviewed and accepted for publication in Proceedings of the 33rd Computer Graphics International, published by ACM on behalf of the author(s). It incorporates referee’s comments but changes resulting from the publishing process, such as copyediting, structural formatting, may not be reflected in this document. The published version is available at: [http://dx.doi.org/10.1145/2949035.2949043].