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

Poster A in Poster Session A: Tuesday, August 4, 9:30 – 11:15 am, Kimmel Center, Shorin & Rosenthal Rooms

Decoding implied motion magnitude using fMRI activity

Koustav Banerjee¹, Daniel Kersten¹, Thomas Naselaris¹; ¹University of Minnesota

Presenter: Koustav Banerjee

While motion sensitive regions MT/MST play a role in the inference of implied motion, how the brain represents the magnitude of implied motion from static scenes remains unclear. To bridge this gap, we employed an LLM to classify images from Natural Scenes Dataset containing implied motion, and also generated a fine-grained motion magnitude score for such images. Using the implied motion classification we conducted a t-test that revealed that higher-level regions including MT, EBA were sensitive to the contrast between scenes with implied motion and scenes without, consistent with previous studies. Next, we found that the implied motion score correlated with the brain activity in the visual cortex, dorsal stream, and anterior regions suggesting that cortical regions show continuous implied motion sensitivity. Finally, we also constructed a linear decoder that could reliably predict the implied motion scores from the brain activity. Overall, our work shows that implied motion from static scenes is represented more broadly through the brain than previously thought, and the brain encodes granular information about the magnitude of motion.

Topic Area: Computational Models of Vision & Visual Cortex