Bathroom

The Bathroom is a typical interior scene with a mix of textured glossy and diffuse materials that is easy to render for a unidirectional path tracer due to large light sources: the windows are modeled as diffuse area emitters. The bump-mapped surfaces are a challenge for local-frame-based guiding methods, such as the GMMs by Vorba et al. [2014] and are handled more gracefully by world-frame-based techniques such as PPG [Müller et al. 2017] and our NPG. In this scene, radiance-based guiding performs worse than unidirectional path tracing due to a rough dielectric glass panel directly in front of the light sources. Product-based techniques, like our NPG-Product, are able to take such BSDF properties into account and are able to significantly improve upon the performance of unidirectional path tracing at equal sample counts. Primary-sample-space path-sampling techniques offer little to no benefit over unidirectional path tracing with our NPS slightly outperforming PSSPS [Guo et al. 2018] with an equal number of samples.
Use mouse wheel to zoom in/out, click and drag to pan. Press keys [1], [2], ... to switch between individual images. Press [?] to see more keybindings.

Charts

Lower values are better (for SSIM, we plot 1-SSIM).
MAPE
SMAPE
L1
MRSE
L2
1-SSIM

Metrics

MetricPath TracingPPGGMMPSSPSNPS KLNPG-Radiance KLNPG-Product KL scalarNPG-Product KLNPG-Product χ²
Mega Samples236236236236236236236236236
Render Time88s2.3m10m89s3.5m9.3m11m12m15m
MAPE0.1470.1890.2720.1430.1460.1780.07140.05420.0655
SMAPE0.1510.1880.2730.1460.1500.1760.07150.05460.0666
L10.09640.1520.2040.09380.09290.1440.05300.04090.0432
MRSE0.03320.05570.1090.03160.03310.05008.89e−34.87e−36.71e−3
L26.84e−30.01240.02226.47e−36.83e−30.01102.08e−31.12e−31.47e−3
1-SSIM0.5710.6380.7370.5590.5670.6200.3010.2120.272