A new per-pixel difficulty descriptor is introduced for heavy-tailed light transport

Summary

The article introduces a discrete transport-mechanism descriptor for per-pixel rendering difficulty; it complements traditional variance-based metrics by classifying light contributions through BSDF lobes and delta-spectral events.

Why it matters

Traditional per-pixel difficulty metrics based on sample variance become unreliable under tail-heavy transport; here, skewness (kurtosis) dominates the error. The authors provide a stable label-based descriptor that remains consistent regardless of sample count, enabling more accurate difficulty assessment and the development of adaptive sampling strategies for complex scenes.

Highlights

• By introducing seven mutually exclusive transport-mechanism labels, it captures the entire energy distribution in tested scenes.
• It demonstrates high label agreement across sample sizes ranging from 87% to 99.6%, surpassing variance-based metrics that can drop as low as 21%.
• It uses the descriptor to correct noisy pilot variance, showing operational gains by improving sample distribution in cases where traditional pilots fail.

Critical take

The descriptor's reliance on detailed event classification may increase computational overhead, and its validation is limited to only seven scenes; broader applicability and integration costs remain uncertain.