We propose a neural approach for estimating spatially varying light selection distributions to improve importance sampling in Monte Carlo rendering, particularly for complex scenes with many light sources. Our method uses a neural network to predict the light selection distribution at each shading point based on local information, trained by minimizing the KL-divergence between the learned and target distributions in an online manner. To efficiently manage hundreds or thousands of lights, we integrate our neural approach with light hierarchy techniques, where the network predicts cluster-level distributions and existing methods sample lights within clusters. Additionally, we introduce a residual learning strategy that leverages initial distributions from existing techniques, accelerating convergence during training. Our method achieves superior performance across diverse and challenging scenes in equal-sample settings.

This project was funded in part by the NSF CAREER Award #2238193. We thank Ryusuke Villemin for the valuable discussions, Vincent Serritella for generating the experimental test data, and Magnus Wrenninge and the Aurora simulation team for support. We are grateful to [Wang et al. 2021] for releasing the source code of their work. We would like to thank the following artists for sharing their scenes and models that appear in our figures: Mareck (Bathroom), SlykDrako (Bedroom), Amazon Lumberyard (Bistro), Jay-Artist (Living Room), Guillermo M. Leal Llaguno (San Miguel), Wig42 (Staircase), NewSee2l035 (Staircase2), and Mike Winkelmann (Zero Day).