Shadow Layers for Participating Media

Abstract

In the movie industry pipeline, rendering programs output the main image along with a collection of Arbitrary Output Variable layers (AOVs) that retain specific information on light transport and scene properties in image space. Compositing artists use AOVs to improve the quality and appearance of the rendered picture during post‐processing, according to the artistic goal of the shot. In particular, cast shadows are manipulated to support narration and storytelling, as the human perception tolerates non‐physical edits. Conventional path tracing renderers often propose a shadow matte AOV containing radiance lost when shadow rays are occluded. Previous work has shown that they incorrectly estimate shadow and miss occluded radiance from indirect light sources, and that shadow layers must be used to correctly recover radiance from single, solid occluders. In this paper, we generalise shadow layers to an arbitrary number of occluders, and add support for participating media. We begin by quantifying the radiance loss between the radiative transfer equation and the rendering equation, and translate it into a path integral formulation for an efficient Monte Carlo integration. We propose a prototype implementation that renders the main image and shadow layers in a single pass with an affordable computational overhead.