When is a closed-form RGB->S/P ratio adequate? A hyperspectral characterization on natural scenes for mesopic display

arXiv cs.GR details a hyperspectral characterization of a closed-form RGB-to-S/P ratio for mesopic display transforms, using natural scenes instead of the narrowband sources that dominated earlier checks. The practical question is whether a low-cost three-channel estimate can stand in for per-pixel spectral S/P when driving low-light rendering and display pipelines.

On a daylight radiance time-series, a six-scalar closed form reproduced spectral S/P with a median error of about 0.07 after chromatic adaptation to D65. In unadapted sRGB, the error drifted with color temperature and landed around 0.19, making adaptation the enabling step for this workflow.

The result held up on an independent 50-scene set, with a pooled median error of 0.024 and 45 of 50 scenes inside a pre-registered 0.10 band. The misses clustered in saturated, spectrally peaky subjects such as floral close-ups, which behave more like the narrowband cases where simple projections break down.

For engine and display developers, the takeaway is that a closed-form RGB proxy can be good enough for broad, natural content, but it is not a universal substitute for spectral data. The scotopic weighting is also model-dependent, though the paper says the chosen value is backed by a primary-free XYZ projection and stays within the same error band across principled coefficient choices.