Fig. 7: Decoding luminance polarity carried by different hues.

a Accuracy of classifiers trained to decode luminance polarity on the identity problems (individual problems, thin lines; average, thick line). b Accuracy of classifiers trained to decode luminance polarity across hues (12 generalizing-across-hue problems, thin lines; average, thick dashed line; shading shows SE across subjects). The thickness of the horizontal line above the x-axis shows the number of decoding problems (maximum in a, 4; maximum in b, 12) that were significant in each 5 ms bin (cluster corrected). c Heatmap showing classifier performance [95% CI, by bootstrapping] for each binary classifier at the initial peak decoding for each problem. Bold numbers show the total number of 5 ms time bins with significant classification performance. The time stamps show when the classifier performance reached the initial peak in decoding accuracy. d Duration of significant decoding of luminance polarity plotted in the cone-opponent color space used to define the stimuli; B, P, O, and G demarcate within this space the location of the blue, pink, orange, and green hues of the stimuli. The color scalebar indicates the number of 5 ms time bins that were significant, and the arrow direction points from the hues used for training to the hues used for testing. For example, the arrow from P to G corresponds to classifiers trained on luminance polarity using data obtained with light/dark pink stimuli and tested using data obtained with light/dark green stimuli. For the four identity problems (the letters B, P, O, and G), the hue was the same for training and testing. For each type of problem, the precise number of significant time bins is shown, along with the [95% CI] obtained by bootstrapping. Problems that cross the color space, in which the hues for training and testing varied in both L-M and S activation, showed lower numbers of significant time bins than problems around the perimeter of the color space, in which the hues for training and testing varied in only L-M or S.