Stochastic rasterization using time-continuous triangles

Abstract

We present a novel algorithm for stochastic rasterization which can rasterize triangles with attributes depending on a parameter, t, varying continuously from t = 0 to t = 1 inside a single frame. These primitives are called time-continuous triangles, and can be used to render motion blur. We develop efficient techniques for rasterizing time-continuous triangles, and specialized sampling and filtering algorithms for improved image quality. Our algorithm needs some new hardware mechanisms implemented on top of today's graphics hardware pipelines. However, our algorithm can leverage much of the already existing hardware units in contemporary GPUs, which makes the implementation fairly inexpensive. We introduce time-dependent textures, and show that motion blurred shadows and motion blurred reflections can be handled in our framework. In addition, we also present new techniques for efficient rendering of depth of field and glossy planar reflections using our stochastic rasterizer.