Face and body representations converge along the visual hierarchy in models and cortex
van Dyck, L. E.; Dobs, K.
Show abstract
Person perception relies on visual cues from faces and bodies, but whether the brain processes these signals separately or jointly remains unclear. Here, we test the extent to which face and body representations emerge as segregated or integrated in deep neural network models, and how these representations map onto human visual cortex. We find that models optimized for visual recognition develop not only face- and body-selective units, but also mixed-selective units responding preferentially to both categories. Using fMRI encoding analyses, we show that mixed-selective units best explain face- and body-selective cortical regions. Face- and body-selective units contribute unique variance in their corresponding regions but predominantly explain shared variance that increased from posterior to anterior cortex. In models, we further demonstrate that mixed selectivity supports multiple person perception tasks but corresponds to part-based processing of whole persons. Our findings reveal progressive integration of face and body representations along the visual hierarchy in both models and cortex.
Matching journals
The top 4 journals account for 50% of the predicted probability mass.
Similar papers in this journal
Similar papers in this journal
- Increasing stimulus similarity drives nonmonotonic representational change in hippocampus 95%
- Stimulus representation in human frontal cortex supports flexible control in working memory 95%
- Successor-like representation guides the prediction of future events in human visual cortex and hippocampus 95%
Similar papers in this journal
- The relative coding strength of object identity and nonidentity features in human occipito-temporal cortex and convolutional neural networks 97%
- The functional organization of high-level visual cortex determines the representation of complex visual stimuli 96%
- Evidence for abstract codes in parietal cortex guiding prospective working memory 96%
"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.