Neural Path Guiding with Distribution Factorization

2025-06-01

AI Summary

Existing neural path-guiding approaches struggle to balance expressive sampling power with computational efficiency. This paper addresses path guiding in Monte Carlo rendering by proposing an efficient and expressive neural directional distribution model. We introduce a novel factorization of the 2D directional distribution into two independent 1D probability density functions (PDFs), enabling fast sampling and exact PDF evaluation. To stabilize gradient estimation and approximate the normalization constant, we incorporate a radiance caching network. Furthermore, we integrate discrete coordinate modeling, bilinear interpolation, and a distribution-similarity maximization training strategy. Experiments demonstrate that our method significantly reduces rendering variance in complex light-transport scenarios, achieves higher image quality than state-of-the-art neural guiding techniques, and maintains high computational efficiency—effectively balancing real-time performance and rendering accuracy.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Search and Optimization: Sampling/Simulation-based Search
• Machine Learning: Learning with Manifolds

Application Category

• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

In this paper, we present a neural path guiding method to aid with Monte Carlo (MC) integration in rendering. Existing neural methods utilize distribution representations that are either fast or expressive, but not both. We propose a simple, but effective, representation that is sufficiently expressive and reasonably fast. Specifically, we break down the 2D distribution over the directional domain into two 1D probability distribution functions (PDF). We propose to model each 1D PDF using a neural network that estimates the distribution at a set of discrete coordinates. The PDF at an arbitrary location can then be evaluated and sampled through interpolation. To train the network, we maximize the similarity of the learned and target distributions. To reduce the variance of the gradient during optimizations and estimate the normalization factor, we propose to cache the incoming radiance using an additional network. Through extensive experiments, we demonstrate that our approach is better than the existing methods, particularly in challenging scenes with complex light transport.

Problem

• Improving Monte Carlo integration in rendering with neural path guiding
• Balancing speed and expressiveness in neural distribution representations
• Enhancing light transport modeling in complex scenes using factorization

Innovation

• Factorizes 2D distribution into two 1D PDFs
• Uses neural networks for discrete PDF estimation
• Caches radiance with network to reduce variance