Mice recognize 3D objects from recalled 2D pictures, support for picture-object equivalence

Abstract

Picture-object equivalence or recognizing a three-dimensional (3D) object after viewing a two-dimensional (2D) photograph of that object, is a higher-order form of visual cognition that may be beyond the perceptual ability of rodents. Behavioral and neurobiological mechanisms supporting picture-object equivalence are not well understood. We used a modified visual recognition memory task, reminiscent of those used for primates, to test whether picture-object equivalence extends to mice. Mice explored photographs of an object during a sample session, and 24 h later were presented with the actual 3D object from the photograph and a novel 3D object, or the stimuli were once again presented in 2D form. Mice preferentially explored the novel stimulus, indicating recognition of the “familiar” stimulus, regardless of whether the sample photographs depicted radially symmetric or asymmetric, similar, rotated, or abstract objects. Discrimination did not appear to be guided by individual object features or low-level visual stimuli. Inhibition of CA1 neuronal activity in dorsal hippocampus impaired discrimination, reflecting impaired memory of the 2D sample object. Collectively, results from a series of experiments provide strong evidence that picture-object equivalence extends to mice and is hippocampus-dependent, offering important support for the appropriateness of mice for investigating mechanisms of human cognition.

Bridging tuning and invariance with equivariant neuronal representations

“…Like primates, rodents can recognize objects using vision alone. [28][29][30] Our experiments show that planar rotation is untangled between V1 and higher-order visual cortical areas in mice. We explain how untangling could arise from visual processing across equivariant layers, and develop a computational method to estimate a rotationequivariant subspace from neuronal responses.…”

Commentary on the scientific rigor of Sen and Gredebäck's simulation: Why empirical parameters are necessary to build simulations

“…The VRM task is identical in principle to the Novel Object or Object Recognition tasks that are used extensively in nonhuman subjects (Ennaceur & Delacour, 1988). Many studies of object recognition in nonhuman animals employ a criterion regarding the amount of time that subjects must interact with, or explore, the object (i.e., a learning criterion) in order for subjects to be included in the analysis of memory (e.g., Cohen et al., 2022; Gaskin et al., 2010; Jablonski et al., 2013; Shimoda et al., 2021). It is also not uncommon in studies of operant conditioning in nonhuman animals (e.g., learning the contingency between some response and an appetitive outcome; akin to the mobile paradigm) for a learning criterion to be applied (e.g., must correctly respond on 80% of choices or responses; Boulougouris et al., 2007; Schoenbaum et al., 1999; Šlipogor et al., 2022).…”