Dynamic context-based updating of object representations in visual cortex

Objects in real-world scenes are often poorly or partially visible, for example because they are occluded or appear in the periphery. An additional challenge of real-world vision is that it is dynamic, causing the appearance of objects (e.g., their size and orientation) to change as we move. Importantly, however, these changes are predictable from the 3D structure of the surrounding scene. In two fMRI studies, we find that visual cortex dynamically updates object representations using this predictive contextual information. Firstly, visual cortical representations of objects were enhanced when they rotated congruently (versus incongruently) with the surrounding scene. Secondly, the inferred orientation of the object could be decoded from visual cortex activity even when the object was fully occluded. These findings indicate that predictive processes in visual cortex follow the geometric structure of the environment, providing a mechanism to support object perception in dynamic natural vision.