Fig. 2: Model performance.
From: Keypoint-based modeling reveals fine-grained body pose tuning in superior temporal sulcus neurons
a Reliability-normalized coefficient of determinations \({\widetilde{{{{\rm{R}}}}}}_{{{{\rm{KP}}}}}^{2}\) for the units for the 2D and 3D_VD keypoint models. Units were ranked according to their predictive performance. The rows and columns of the panels are for the monkeys and regions, respectively. The shadowed regions correspond to percentiles 2.5 and 97.5 of the null distributions of \({\widetilde{{{{\rm{R}}}}}}_{{{{\rm{KP}}}}}^{2}\) estimated using permutation of the stimulus labels (N = 1000). The horizontal arrows correspond to the median \({\widetilde{{{{\rm{R}}}}}}_{{{{\rm{KP}}}}}^{2}\) for 2D and 3D_VD models, per monkey and region. b Distribution of differences between adjusted reliability-normalized coefficients of determination, denoted as \({\widetilde{\widetilde{{{{\rm{R}}}}}}}^{2}\), between the keypoint-based model and AlexNet for different layers of the AlexNet. P values from two-sided Wilcoxon signed rank tests. Columns and rows correspond to regions and monkeys, respectively. n indicates the number of units. Box plots show the median (horizontal line), interquartile range (box: 25th–75th percentile), and data within 1.5× the interquartile range from the lower and upper quartiles (whiskers). Points beyond this range are plotted individually as outliers. Source data are provided as a Source Data file.
