Selective guided sampling with complete light transport paths

Abstract

Finding good global importance sampling strategies for Monte Carlo light transport is challenging. While estimators using local methods (such as BSDF sampling or next event estimation) often work well in the majority of a scene, small regions in path space can be sampled insufficiently (e.g. a reflected caustic). We propose a novel data-driven guided sampling method which selectively adapts to such problematic regions and complements the unguided estimator. It is based on complete transport paths, i.e. is able to resolve the correlation due to BSDFs and free flight distances in participating media. It is conceptually simple and places anisotropic truncated Gaussian distributions around guide paths to reconstruct a continuous probability density function (guided PDF). Guide paths are iteratively sampled from the guided as well as the unguided PDF and only recorded if they cause high variance in the current estimator. While plain Monte Carlo samples paths independently and Markov chain-based methods perturb a single current sample, we determine the reconstruction kernels by a set of neighbouring paths. This enables local exploration of the integrand without detailed balance constraints or the need for analytic derivatives. We show that our method can decompose the path space into a region that is well sampled by the unguided estimator and one that is handled by the new guided sampler. In realistic scenarios, we show 4× speedups over the unguided sampler.

A practical path guiding method for participating media

“…Reibold et al [4] proposed using path guiding only for paths with high variance or difficult paths, and regular path tracing for other paths, as path guiding is more expensive compared to regular path tracing.…”

“…Path guiding has been recently introduced for surface rendering [1][2][3][4]. The common goal of these works is to find an "optimal" distribution that can approximate the actual path integral to make convergence faster.…”

Practical Product Path Guiding Using Linearly Transformed Cosines

“…However, the training step is computationally expensive and it can be hard to evaluate how much training is required, i.e., whether the current distributions are sufficiently converged. Reibold et al [RHJD18] propose the use of an outlier rejection algorithm to determine paths in a scene with high variance and apply guiding using GMMs only to those paths. In this way they use expensive guiding only where necessary.…”

“…all guiding methods. Despite recent work (e.g., [RHJD18,BJ19]), there is currently no easy way to identify "difficult paths" for which product guiding is guaranteed to be cost effective.…”

“…However, the treatment of such light interactions, e.g., the decision to reflect or refract could result in paths with high contribution where guiding could be beneficial. It is unclear how to adapt our data structures to effectively guide such sampling decisions, without storing the complete path [RHJD18].…”