Table 3.

Average relative reduction in prediction error (mean $\pm$ SD) for fits of the general 12-parameter Affine map from the original convergence model papers to our stimuli, broken down according to color space and amount of simulated refraction. Values are averaged over mask type, illuminant color, transparent body color, whether the fit was done for the full-color distribution or chromaticity only, and whether or not dark pixels were excluded. The relatively low standard deviation in all cases indicates that variations in all of these factors play relatively little role in the applicability of the convergence model to our stimuli. See the main text for more details.

	LMS	MB-DKL	CIELAB
No refraction	35.8 ± 8.74%	42.6 ± 18.8%	51.1 ± 13.0%
Little refraction	42.2 ± 7.76%	44.6 ± 18.5%	46.9 ± 15.1%
Default glass refraction	56.9 ± 14.8%	57.3 ± 17.2%	49.0 ± 13.7%