Rendering with Coherent Layers

Published by Association for Computing Machinery, Inc.

Publication

For decades, animated cartoons and movie special effects have factored the rendering of a scene into layers that are updated independently and composed in the final display. We apply layer factorization to real-time computer graphics. The layers allow targeting of resources, whether the ink and paint artists of cartoons or the graphics pipeline, to those parts of the scene that are most important. To take advantage of frame-to-frame coherence, we generalize layer factorization to apply to both dynamic geometric objects and terms of the shading model, introduce new ways to trade off fidelity for resource use in individual layers, and show how to compute warps that reuse renderings for multiple frames. We describe quantities, called fiducials, that measure the fidelity of approximations to the original image. Layer update rates, spatial resolution, and other quality parameters are determined by geometric, photometric, visibility and sampling fiducials weighted by the content author’s preferences. We also compare the fidelity of various types of reuse warps and demonstrate the suitability of the affine warp. Using Talisman, a hardware architecture with an efficient layer primitive, the work presented dramatically improves the geometric complexity and shading quality of scenes rendered in real-time.