← All resources

Neural Importance Sampling of Many Lights

Auto-discovered

Paper Advanced

Pedro Figueiredo, Qihao He, Steve Bako, Nima Khademi Kalantari

Visit resource ↗

Description

Neural Importance Sampling of Many Lights presents a SIGGRAPH 2025 Conference Papers method by Pedro Figueiredo, Qihao He, Steve Bako, and Nima Khademi Kalantari for learning spatially varying light selection distributions in many-light Monte Carlo rendering.

A neural network predicts light selection from local shading features by minimizing KL divergence to a target distribution online. For hundreds or thousands of lights, the network scores hierarchy clusters while conventional methods sample inside a cluster, and residual learning warms starts from existing hierarchy distributions to speed convergence. Equal-sample results are reported across challenging many-light scenes.

It is for advanced rendering researchers and path-tracing engineers who already use light trees or many-light sampling and want learned selection on top.

Access is a project page with abstract, author affiliations (Texas A&M and Aurora Innovation), acknowledgements, and BibTeX for the ACM paper.

It is useful when evaluating neural light selection that stays compatible with hierarchical many-light pipelines. Link last verified accessible on 2026-09-29. Link last verified accessible on 2026-09-28.