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

Poster B in Poster Session B: Tuesday, August 4, 2:00 – 3:45 pm, Kimmel Center, Shorin & Rosenthal Rooms

Characterizing and modeling time-varying object percepts evoked by naturalistic video

Lynn K. A. Sörensen¹, Michael J. Lee¹, Aicha Masmoudi¹, Nancy Kanwisher¹, James J. DiCarlo¹; ¹Massachusetts Institute of Technology

Presenter: Lynn K. A. Sörensen

How do our brains continuously construct and update rich perceptual states from dynamic sensory input? While computational models can now explain how humans perceive objects in images, we lack an analogous understanding of how sensory history modulates object perception. Here, we make two contributions: (1) a high-throughput online psychophysics paradigm that quantifies human perceptual states in continuously unfolding everyday scenes (0.1–15s), yielding highly reliable reports across 12 object categories and implying that observers build time-varying and persistent perceptual states that extend beyond what is currently visible; and (2) initial computational models combining frame-based vision encoders with transformer-based attentional pooling. Our results demonstrate that models that integrate high-level visual representations over time best account for human object perception across timescales of up to 15s.

Topic Area: Computational Models of Vision & Visual Cortex