Figure 3
From: Object geometry serves humans’ intuitive physics of stability
Stimuli and results for Experiments 1 and 2. (A) The eleven objects used across experiments. Color patches correspond to objects in the data panels (B and C). (B) Point of subjective equality (PSE) estimates of critical angles plotted for adult and child participants in each experiment. (C) PSE estimates plotted for each of the geometric models and ResNet-50 in each experiment. Error bars represent 95% bootstrapped CIs. Adults and children performed comparably across all experiments, except in Experiment 2B in which children did not show differences in PSE estimates between the two objects (whereas adults did). Among the geometric models, the centroid model showed performance that was qualitatively similar to human participants across all conditions. All other geometric models differed from humans in at least one of the conditions by displaying opposite response profiles. ResNet-50 showed qualitatively similar response profiles to humans, except in Experiment 2A.
