Space-Optimized Texture Atlases for 3D Scenes with Per-polygon Textures

Real-time rendering of massively textured urban scenes is still a challenging problem in computer graphics. Texture atlases and the recently-introduced texture arrays have proven to be a powerful tool to accelerate the rendering of such scenes by minimizing the number of texture switches. In this paper we present a perception-based scheme for generating space-optimized texture atlases from 3D scenes with per-polygon textures. Key elements of our approach include the use of perceptual metrics to allocate texture space in accordance to the visual content of each chart, a joint encoding of chart size and position coordinates to provide within-chart tiling support with no space overhead, and a packing strategy achieving full occupancy of the texture atlas. We demonstrate our optimization scheme in the context of real-time navigation through a gigatexel urban model. Our scheme significantly decreases texture memory usage and provides dramatic speed-up factors without noticeable loss in visual quality.