Fig. 2: Summary of behavioural results.
From: Human gloss perception reproduced by tiny neural networks
a, Each data point shows a specific stimulus image, and the plot includes data on all 3,888 images. The horizontal and vertical axes show physical ground-truth reflectance values and observer settings (perceived gloss) averaged over observers. b, Histogram of human versus ground-truth correlation. Corr., correlation; obs, observer; gt, ground truth. The plot shows the distribution of Pearson’s correlation coefficients between each observer’s settings and the corresponding ground-truth values. c, Histogram of within-observer correlations. For each observer, Pearson’s correlation coefficient was computed between session 1 and session 2 to quantify the intra-observer variability. d, Histogram of correlation across observers. For the left plot, the correlation was computed between one observer and the average across all other observers across 72 images. For the right plot, the correlation was computed over 12 images that were shared across all 54 image sets for every pair of observers (that is, (295 × 294)/2 = 43,365 pairs). e, Each grid shows eight example images. The horizontal and vertical positions in the grid show the physical specular reflectance assigned to each object and observer settings (perceived gloss level). The number of images in each grid is shown at the right bottom of each grid.
