Parallel construction of object motion from retina to cortex

Abstract

Detecting a moving object against a moving background is fundamental to survival. The retina is known to perform this computation, extracting object motion from global motion before signals reach the brain. Under the prevailing hierarchical view of vision, downstream areas should inherit and elaborate this retinal computation rather than rebuild it. Using multielectrode recordings of matched stimuli in mouse retina and cortex along with computational modeling, we show instead that post-retinal processing recomputes rather than inherits object motion sensitivity. Retinal and cortical object-motion sensitivities have distinct tuning: the retina prefers fine jitter, whereas cortex prefers coarse drift, revealing a division of labor in which the retina detects moving objects, whereas the cortex conveys information about object pattern. Vision synthesizes object motion through parallel specialization, not hierarchical refinement.