Object size and depth representations in human visual cortex

One of the key abilities in human object perception is maintaining a reliable representation of an object’s real-world size across various distances and perspectives. Previous research has indicated that neural responses in the ventral temporal cortex reflect object real-world size representations. However, the stimuli used in these prior studies confounded two related properties: perceived real-world size and real-world depth (distance). Moreover, the stimuli did not include naturalistic backgrounds, preventing us from exploring visual mechanisms in more ecological conditions. Bridging this limitation, a recent study from our group conducted a model-based representational similarity analysis on EEG data from a large-scale dataset of subjects viewing natural images featuring objects of varying retinal sizes and depths. The EEG study successfully disentangled a distinct timeline of processing objects real-world size and real-world depth. To better understand object representations in human brain regions with better spatial resolution, our current study applies a similar analysis approach to fMRI data, aiming to explore how different parts of human visual cortex represent objects real-world size and depth information in natural images. Applying our model-based representational similarity analysis on the THINGS fMRI dataset, we isolated neural representations specific to real-world size, real-world depth, and retinal size across human visual cortex. We found the most robust real-world depth representations in scene-selective regions such as the Parahippocampal Place Area (PPA) and the Transverse Occipitial Sulcus (TOS), and the most robust real-world size representations in middle-level visual regions, such as the V4, V3A and V3B. Our study delineates how various regions in human visual cortex are involved in processing different object size and depth features via an advanced computational approach, which offers an insightful understanding of the human brain processing of object information within naturalistic images.