Author summary Understanding the origins of shape perception is a long-standing challenge in psychology and neuroscience. We bring together methods from developmental psychology, computational neuroscience, virtual reality, and artificial intelligence to characterize the learning mechanisms, experiences, and sensors for developing shape perception. We find that shape perception develops from three ingredients: (1) generic fitting systems that operate by the same general fitting principles as evolution, (2) embodied visual experiences that include diverse temporally linked views of objects, and (3) biologically plausible sensors that reformat raw visual input like the human retina. When artificial agents are equipped with generic fitting systems (brains) and human-like artificial retinas (eyes), the agents automatically develop a shape-centric visual system that categorizes novel objects by shape. Our results provide an image-computable origin story for shape perception.
Introduction
Shape perception is central to human vision [1–4]. We perceive the world in terms of bounded, cohesive objects that maintain their shape across space and time [5]. We prioritize shape over other cues (e.g., size, color, texture) when rec… [65068 chars]
Source: PLOS (Public Library of Science) | Published: 2025-12-15T00:00:00Z
Credit: PLOS (Public Library of Science)
