Warts, cracks, causal shape segmentation.
Research on shape perception usually focuses on the estimation of local surface geometry through cues like stereopsis, shading, or texture. Here, we argue that observers use these shape estimates to infer other object properties such as material composition and the transformation processes that generated the observed shape from this matter. We call this separation of object shape into intrinsic and extrinsic object properties shape scission. We investigated shape scission in a series of experiments with different groups of participants responding to a set of 8 unfamiliar rendered objects, each transformed by 8 transformations (e.g., “melted”, “cut”, or “inflated”). Importantly, participants could never directly compare the transformed and untransformed versions of objects.
First, participants produced adjectives in a free naming task to describe what happened to the transformed objects. Second, other participants classified the objects according to either (1) their original shape, or (2) the transformation that had been applied to them. Third, participants marked those regions of the objects that were transformed away from the original shape. Finally, participants viewed objects at 5 different levels of transformation magnitude and provided perceptual ratings of deformation. We find that participants (i) are consistent in naming the transformations, (ii) can classify unfamiliar objects according to their original shape as well as the applied transformation, (iii) can identify regions of the objects that were transformed, and (iv) can to some extent perceive the magnitude of the transformation (when compared to objective mesh deformations). Thus, we can identify “objects” across transformations and “transformations” across objects, separating observed features by their causal origin (shape scission).This perceptual understanding of causal processes allows us to infer not only how objects have been altered by forces in their past but also what other members of the same class might look like.
Schmidt, Phillips & Fleming 2018