This is the authors' abstract. We don't add key points for this paper.
Human competencies in object recognition and category generalization implicate representations of the 3D geometry of objects. There is no accepted account of how such object-centric representations are formed. Toddlers' manual and visual exploration of objects is strongly associated with their emerging abilities in object recognition and category generalization, although there is limited understanding of the processes that underlie this association. Building on Marr and Nishihara's seminal theory of shape, we proposed (1) that toddlers actively select views that optimize fundamental visual sensitivities for long extents and for edges oriented horizontally and vertically and (2) that these biased views result in the axis of the object's maximum elongation, the major axis, being stably present in the variable 2D shapes projected from the object to the eye. We analyzed the visual properties of toddlers' (n = 90, 12-38 months) egocentric views as they manipulated objects. We observed a highly biased sampling of object views in which the major axis of the physical object is near maximal extension and oriented near horizontal or vertical, two properties that make the major axis stably determinable in the variable 2D shapes projected to the eye. The findings show how fundamental low-level visual sensitivities may play a critical developmental role in biasing behaviors that bias the input in ways that make hard computational problems tractable. The findings also provide an empirical agenda for research focused on the development of object-centric representations of 3D shape. SUMMARY: We examined the visual properties of toddlers' egocentric views of objects. We found that the generated views were a selective sampling of possible views that optimized foundational low-level visual properties for shape perception. The findings implicate strong intrinsic constraints on toddlers' visual experiences of object views. The biases in the toddler generated views are consistent with class theories of object-centric representations of 3-dimensional geometry.
Answers are grounded in this paper and cite their sources.
No takes yet. Share an insight, caveat, or question.
Pereira et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: