Rendering Participating Media Using Path Graphs

TL;DR

This paper extends the path graph framework to participating media by introducing volume-specific aggregation and propagation operators that reuse information from multiple-scattering paths to accelerate convergence. The method reformulates the volume radiative transfer problem with linear operators and MIS-based aggregation across spatial clusters, enabling global information sharing beyond local neighborhoods. The authors demonstrate substantial variance reduction and faster convergence in challenging heterogeneous and forward-scattering media, implemented via a CPU path tracer and a GPU-accelerated path-graph solver. The approach improves volumetric rendering realism and speed, offering a practical pathway for faster design iterations in scenes with clouds, fog, smoke, and subsurface scattering.

Abstract

Rendering volumetric scattering media, including clouds, fog, smoke, and other complex materials, is crucial for realism in computer graphics. Traditional path tracing, while unbiased, requires many long path samples to converge in scenes with scattering media, and a lot of work is wasted by paths that make a negligible contribution to the image. Methods to make better use of the information learned during path tracing range from photon mapping to radiance caching, but struggle to support the full range of heterogeneous scattering media. This paper introduces a new volumetric rendering algorithm that extends and adapts the previous \emph{path graph} surface rendering algorithm. Our method leverages the information collected through multiple-scattering transport paths to compute lower-noise estimates, increasing computational efficiency by reducing the required sample count. Our key contributions include an extended path graph for participating media and new aggregation and propagation operators for efficient path reuse in volumes. Compared to previous methods, our approach significantly boosts convergence in scenes with challenging volumetric light transport, including heterogeneous media with high scattering albedos and dense, forward-scattering translucent materials, under complex lighting conditions.

Figures (6)

• Figure 1: Our method fits between the traditional steps of path tracing and denoising. This means that it can provide additional benefits on top of (rather than replacing) techniques like neural denoisers or advanced path sampling. Unlike the surface Path Graph, the final gather is optional in the volume case. The final gather serves to remove correlations between neighboring pixels, which are very subtle in volumes since path vertices are scattered in the volume.
• Figure 2: Illustration of the volume path graph construction from paths sampled during a standard volumetric path tracing pass with next event estimation in participating media.
• Figure 3: We show an equal sample (at 1 sample per pixel) comparison between ours and path tracing. Our method extensively reuses the paths' information even with only one sample per pixel, significantly improving rendering efficiency.
• Figure 4: Comparison between (a) bidirectional path tracing, (b) photon mapping, (c) path tracing and (d) our method (path tracing + path graphs) on an equal time (5 min) in scenes with the presence of homogeneous volume. We show full light transport in all the scenes and our method provides significant variance reduction over previous methods on rendering the participating media. The $2^{\text{nd}}$ and $3^{\text{rd}}$ scenes contain large volume of outdoor medium (e.g. fog), where (unguided) photon mapping suffers since only a small amount of photons arrive at the region where the camera is looking towards due to strong multiple scattering. The $2^{\text{nd}}$ scene has a skydom as environment lighting as well as the traffic lights, where the bidirectional method failed to sample the small light sources.
• Figure 5: Comparison between (a) path tracing and (b) our method (path tracing + path graphs) on an equal time (5 min) in scenes with the presence of heterogeneous volume. We show full light transport in all the image and our method out perform the path tracing especially around the area where the pixel value is dominated by multi-bounces.