Predicting upcoming visual features during eye movements yields scene representations aligned with human visual cortex.

GPNs predicting upcoming visual features during eye movements produce scene representations that strongly align with human visual cortex responses.

• Why it matters: Understanding how the brain extracts complex scene information during natural vision addresses a key gap in modeling human visual processing and may inform biologically plausible AI systems.
• What they did: Researchers developed glimpse prediction networks (GPNs), recurrent models trained to predict the embedding of the next glimpse along human-like scanpaths, capturing object co-occurrences and spatial arrangements.
• The result: GPN representations match or outperform alternative models in aligning with human fMRI responses in mid- and higher-level visual cortex, supporting next-glimpse prediction as a biologically plausible approach.