← All resources

Practically Utilizing Neural Networks in CPU-based Production Rendering

Auto-discovered

Paper Advanced

Steve Bako, Ryusuke Villemin, Magnus Wrenninge

Visit resource ↗

Description

Practically Utilizing Neural Networks in CPU-based Production Rendering presents a Journal of Computer Graphics Techniques paper by Steve Bako, Ryusuke Villemin, and Magnus Wrenninge (Aurora Innovation) on embedding lightweight neural nets inside CPU path tracers.

It explains why per-sample, per-bounce inference resists typical GPU batching, then describes offline-trained MLPs deployed as frozen weights with Eigen and ISPC on the CPU. Two production cases are detailed: specular BRDF energy compensation and atmospheric or skydome radiance evaluation, with gains in memory, speed, and quality versus LUT-heavy baselines.

The audience is advanced production rendering engineers on CPU VFX pipelines. Familiarity with path tracing and microfacet BRDFs is assumed.

The JCGT page offers full-text and low-resolution PDFs, BibTeX, and a supplemental ZIP under open access.

It shows a practical path to ML inside the render loop without forcing GPU inference into CPU farm workflows. Link last verified accessible on 2026-09-29. Link last verified accessible on 2026-09-28.