Abstract
Real-time Rendering with Wavelet-Compressed Multi-Dimensional Datasets on the GPU
by: Stephen DiVerdi, Nicola Candussi, and Tobias Hollerer
Abstract:
We present a method for using large, high dimension and high dynamic range datasets on modern graphics hardware. Datasets are preprocessed with a discrete wavelet transform, insignificant coefficients are removed, and the resulting compressed data is stored in standard 2D texture memory. A set of drop-in shader functions allows any shader program to sample the wavelet-encoded textures without additional programming. We demonstrate our technique in three applications - a terrain renderer with a 16384^2 sample RGB texture map, a volume renderer with a 512^3 sample 3D dataset, and a complex material shader using an unapproximated BRDF dataset, sampled at 64^4 in RGB.
Keywords:
Texture Compression, Wavelets, Programmable Shaders, Large Textures, BRDF, Volume Rendering, Multi-dimensional Functions
Date:
January 2005
Document: 2005-05