Abstract
Objects can be recognized from visual features at multiple spatial scales, from coarse global shape to fine local detail. How infant visual cortex encodes these features across the visual hierarchy, however, remains unknown. We combined awake infant fMRI with developmentally inspired deep neural networks (DNNs) trained and tested with different levels of image blur to probe cortical tuning to coarse and fine visual features. V1 showed mature tuning to coarse and fine features by 2 months. In contrast, infant extrastriate cortex (V2-V4) followed a protracted developmental trajectory and was biased toward coarse object features relative to adults. This relative bias persisted at 9 months and was not readily explained by developing acuity. The sustained coarse-feature bias contrasted with DNNs, in which low-spatial-frequency representations took longest to converge with visual cortex during training, revealing distinct developmental biases in object-feature tuning in biological and artificial visual systems.