Fig. 6
Curvature modification does not explain blur selectivity. a Schematic of analysis performed across shape and blur datasets to assess the contribution of curvature modification toward blur selectivity. An APC model is fit to data collected from shape screening (red), which then predicts responses to modified curvature threshold contours computed from blurred stimuli at different thresholds (blue). b A blurred stimulus (β = 0.32) generated from a shape contour (red) and a family of closed contours defined by the level set of an intensity threshold (blue), each with reduced curvature magnitudes. c For each cell, the minimum prediction error across all intensity thresholds (threshold curvature NRMSE; see Results) plotted as a function of a blur-invariant mean model’s prediction error. The latter predicts responses to different shapes in accordance with the APC model ignoring blur; responses are identical overall blur levels for a given shape. Example cells are filled and labeled. d Threshold prediction error as a function of bootstrapped training error, i.e., a baseline estimate at how well the APC model predicts shape data
