Ashley Au¹ (), Marvin Chun, Yaoda Xu; ¹Yale University

Object recognition is an integral part of the visual system that allows humans to extract different physical characteristics of the objects around them. Regions of the macaque inferotemporal (IT) cortex have shown to exhibit differential preferences for objects based on their spikiness/stubbiness (Bao et al., 2020). The same spiky/stubby distinction is also present in convolution neural networks (CNN) trained for object classification, suggesting that this is likely a fundamental feature dimension in visual object representation (Yargholi & Op de Beeck, 2022). However, similar findings have not been observed in human fMRI studies (Yargholi & Op de Beeck, 2022; Coggan & Tong, 2023). The aim of this study is to reexamine this discrepancy by creating a better-controlled inanimate stimulus set and by testing whether the spikiness/stubbiness distinction influences visual search performance. Based on the output from a CNN previously shown to correspond to macaque IT cortex in its representations of spiky and stubby objects, we selected 6 pairs of spiky and stubby objects matched in semantic category, with each object containing 8 different exemplars. We asked human participants to search for the presence of a target object exemplar among the exemplars of a distractor object. Across two experiments using either intact object images (n = 13) or images equated in low-level features (e.g., luminance, contrast, spatial frequency; n = 14), we observed faster search speeds between the spiky and stubby objects (M = 631 and 977 ms for both experiments, respectively) than within spiky or within stubby objects (M = 724 and 1148 ms; differences of within and between, ps < .001). These results are consistent with spikiness/stubbiness being a salient dimension in visual object representation. Next, we will collect fMRI responses from the human IT cortex for these objects and examine whether a similar saliency is present in neural responses.

Acknowledgements: Grant support: YX is supported by NIH grant 1R01EY030854.