Search Papers | Poster Sessions | All Posters
Poster A123 in Poster Session A - Tuesday, August 6, 2024, 4:15 – 6:15 pm, Johnson Ice Rink
Algorithmic and architectural constraints on human 3D visual inferences
tyler bonnen1 (), Anthony Wagner2, Daniel Yamins2; 1University of California, Berkeley, 2Stanford University
Perception unfolds across multiple timescales. Ventral temporal cortex (VTC) supports visual inferences that are possible 'at a glance' (i.e.<200ms), such as object classification. Other visual inferences, such as inferring the 3D shape of unfamiliar objects, require more time. Using a combination of psychophysics, electrophysiological, and lesion data, here we identify neural structures and algorithms that underlie this ability. First, we compare an online cohort of human participants to electrophysiological recordings from macaque VTC. While performance 'at a glance' is predicted by VTC responses, humans outperform VTC with increased stimulus viewing time. Next, we demonstrate that a neural system downstream of VTC, medial temporal cortex (MTC), plays a causal role in these temporally extended visual inferences. Finally, through a series of in lab eyetracking experiments, we demonstrate that sequential visual sampling of object features is both reliable across participants and necessary for performance. From these data, we suggest that MTC support visual inferences by integrating over visuospatial sequences, providing algorithmic and architectural constraints for theories and models of human perception.
Keywords: ventral temporal cortex medial temporal cortex 3D vision temporal dynamics