Object-selective cortex incorporates predictions from scene context to aid object recognition

Our ability to recognize objects is strongly facilitated by scene context. Scene context does not only predict the identity of an object (a road is likely to contain a car), but also its appearance (e.g., a car is about as wide as the lane on which it’s driving). Using a multi-method approach, we investigated how object processing is influenced by the predicted (retinal) size of an object, as inferred from the current viewing distance. To this end, we created a large set of outdoor scenes, each containing one object, positioned relatively nearby in one condition (thus producing a relatively large retinal image) and relatively far away in the other (producing a relatively small retinal image). For each scene, we additionally created two conditions with coherently sized objects, by swapping the positions of the ‘large’ and ‘small’ objects in the scene. A series of behavioral experiments showed that coherently sized objects are more easily recognized than incoherently sized objects. This recognition advantage was severely reduced when removing global scene information. Using fMRI searchlight analyses, we found that activity patterns evoked by different object categories were more dissociable (by a classifier trained on isolated objects) when objects were coherently compared to incoherently sized. This sharpened representation of size-coherent objects was specific to object-selective cortices, lateral occipital complex (LOC) and posterior fusiform sulcus (pFs), and predicted the magnitude of the behavioral recognition advantage (on an image-by-image basis) in the behavioral experiment. Using online chronometric TMS, we found that stimulating LOC from 160ms after scene onset reduced this behavioral effect to baseline. This establishes a causal role of object-selective cortex in instantiating this viewpoint-dependent behavioral recognition advantage. We conclude that the predicted real-world size of an object, as inferred from scene context, contributes to object recognition and sharpens object representations in object-selective visual cortex.