Schedule of Events | Search Abstracts | Invited Symposia | Symposia | Poster Sessions | Data Blitz

Neural responses in the macaque inferotemporal cortex to translucent-object images

Poster Session A - Saturday, March 7, 2026, 3:00 – 5:00 pm PST, Fairview/Kitsilano Ballroom
Also presenting in Data Blitz Session 4 - Saturday, March 7, 2026, 10:30 am – 12:00 pm PST, Salon F.

Hoko Nakada¹ (), So Fujimoto², Takeshi Matsuo³, Keisuke Kawasaki⁴, Shin'ya Nishida⁵, Ryusuke Hayashi¹; ¹Natl. Inst. of Advanced Industrial Sci. and Technol. (AIST), Tsukuba, Japan, ²Mitsui Mem. Hosp., Tokyo, Japan, ³Tokyo Metropolitan Neurolog. Hosp., Tokyo, Japan, ⁴Niigata Univ. Med. Sch., Niigata, Japan, ⁵Kyoto Univ., Kyoto, Japan

The appearance of translucency is a key visual cue for inferring an object’s material, yet the neural substrate supporting its processing remains unknown. Here, we recorded neural activity from the macaque inferotemporal (IT) cortex, spanning TEO to TE, in response to images of objects varying in translucency. Images were generated by three different methods: (i) images produced by an unsupervised generative model trained on natural object images, systematically varying in both object shape and levels of translucency (grayscale; mean luminance equalized); (ii) scrambled control images produced by randomizing pixels within each object silhouette of (i); and (iii) computer-generated (CG) images with translucent objects rendered by manipulating multiple optical parameters. We obtained human subjective ratings of transparency for each image and directly compared these ratings with macaque IT responses. Representational dissimilarity matrices (RDMs) of IT responses for model-generated images showed significant correlations with the RDM derived from human ratings and exceeded those obtained for scrambled images. Furthermore, a linear discriminant analysis (LDA) classifier trained on IT responses to model-generated images predicted human ratings and generalized to the CG-rendered images, some of whose rating values were outside the range covered by the model-generated set. Both RDM and LDA analyses suggested shape-independent coding of translucency, particularly in posterior IT. Our findings show that macaque IT population activity encodes information related to the appearance of translucency that aligns with human perception, with some sites exhibiting shape-independent responses. These patterns are neither reducible to low-level image features (e.g., mean luminance) nor contingent on image-generation methods.

Topic Area: PERCEPTION & ACTION: Vision