Adaptive Supersampling for Deferred Anti-Aliasing
Résumé
We present a novel approach to perform anti-aliasing in a deferred-rendering context. Our approach is based on supersampling; the scene is rasterized into an enlarged geometry buffer, i.e., each pixel of the final image corresponds to a window of attributes within this buffer. For the final image, we sample this window adaptively based on different metrics accounting for geometry and appearance to derive the pixel shading. Further, we use anisotropic filtering to avoid texturing artifacts. Our approach concentrates the workload where needed and allows faster shading in various supersampling scenarios, especially when the shading cost per pixel is high.