Scenes

atelier

bedroom

bedroom - Isik21 Denoiser

kitchen

sanmiguel

This supplemental material contains the validation videos of each scene. The videos consist of a section with static lighting and moving camera, then static camera with rotating envmap, and finally moving camera and rotating envmap simultaneously. For each scene we provide individual videos and side-by-side comparisons of:

(1) A ground truth reference in Falcor [Benty20].
(2) The four proposed architectures (baseline, prt-inspired, albedo-factorized and diffuse-specular-split).
(3) Real-time path tracing (at 5spp), and the Optix denoiser output.

Additionally, our comparisons include:
(1) The Neural Radiance Cache [Mueller21], full method and cache only, for the scenes atelier and bedroom.
(2) DeepShading's network architecture [Nalbach17], reduced to equal number of parameters and re-trained per scene on atelier and bedroom.
(3) An indicative result of the most recent denoiser [Isik21] on the bedroom scene in a different rendering engine.

References:

BENTY N., YAO K.-H., CLARBERG P., CHEN L., KALLWEIT S., FOLEY T., OAKES M., LAVELLE C., WYMAN C.: The Falcor rendering framework. August 2020. https://github.com/NVIDIAGameWorks/Falcor

MULLER T., ROUSSELLE F., NOVAK J., KELLER A.: Real-Time Neural Radiance Caching for Path Tracing. SIGGRAPH 2021.

NALBACH O., ARABADZHIYSKA E., MEHTA D., SEIDEL H.-P., RITSCHEL T.: Deep Shading: Convolutional Neural Networks for Screen-Space Shading. EGSR 2017.

ISIK M., MULLIA K., FISHER M., EISENMANN J., GHARBI M.: Interactive monte carlo denoising using affinity of neural features. SIGGRAPH 2021.