Keynotes | K&Ts | GACs | Talks | Posters | Search

Poster E in Poster Session E: Thursday, August 6, 10:30 am – 12:15 pm, Kimmel Center, Shorin & Rosenthal Rooms

Evidence for Object-centric Association Fields in Human Visual Cortex

Hossein Adeli¹, Seoyoung Ahn², Andrew F. Luo³, Mengmi Zhang⁴, Gregory J. Zelinsky⁵, Nikolaus Kriegeskorte¹; ¹Columbia University, ²Hankuk University of Foreign Studies, ³University of Hong Kong, ⁴Nanyang Technological University, Singapore, ⁵State University of New York at Stony Brook

Presenter: Hossein Adeli

Object-based feature grouping in natural scenes requires the visual system to bind image regions into coherent wholes despite ambiguous boundaries and overlapping objects. Association fields in early visual areas have been shown to link neurons with similar low-level features, but grouping in natural scenes requires richer, object-centric connectivity in visual areas. Here we provide evidence for the existence of such object-centric association fields in human visual cortex. Using the Natural Scenes Dataset, a large-scale fMRI dataset of responses to natural scenes, we show that models with stronger object-centric local feature similarity structure better predict brain responses in visual areas. Supervised models fine-tuned to match the Gram matrix structure of a self-supervised model show corresponding improvements in brain predictivity in these areas, pointing to local feature correlation structure as the active ingredient. Together, these results provide evidence for object-centric association fields in human visual cortex and point to feature correlation structure across local representations as a key organizing principle of object-level processing in the brain.

Topic Area: Computational Models of Vision & Visual Cortex