Tien Dung Nguyen¹ (tien-dung.nguyen@utexas.edu), Yoon H. Bai², Corey M. Ziemba³, Olivier J. Hénaff⁴, Robbe L. T. Goris¹; ¹The University of Texas at Austin, ²Massachusetts Institute of Technology, ³National Eye Institute, ⁴Cursive, London, UK

Primates rely on their visual system to infer current and future states of the environment. Natural visual input typically evolves along complex temporal trajectories that are difficult to extrapolate. Recent perceptual experiments in humans and physiological experiments in monkeys suggest that the visual system transforms natural video trajectories into straighter neural representations, thereby facilitating temporal prediction. While both types of experiments yielded qualitatively similar results, perceptual effects were substantially stronger. Which factors underlie this discrepancy? Here we tested two hypotheses. First, that the discrepancy is due to differences in stimulus location. Stimuli in the perceptual study overlapped with both the central and peripheral visual fields, whereas stimuli in the physiological study were limited to the periphery. If perceptual straightening varies with eccentricity, this could account for the discrepancy. We measured perceptual curvature of 6 video sequences presented in the fovea, parafovea, and periphery and found that natural videos elicit specific and robust straightening effects across the visual field (median change in curvature: fovea: -45.7°, parafovea: -58.8°, periphery: -40.4°). Alternatively, the discrepancy may be due to differences in neural processing stages. Perceptual experiments engage many stages of the visual hierarchy. In contrast, the physiological experiments were exclusively conducted in the primary visual cortex (V1). Perceptual straightening may arise from a cascaded computation that also involves later processing stages. To test this hypothesis, we conducted new electrophysiological experiments in two awake, fixating macaques. We measured the curvature of neural population trajectories elicited by 16 videos in the first and second visual area (V2). We found that natural videos elicit specific neural straightening effects that increase in strength between V1 and V2 (median change in curvature: V1: -2.7°, V2: -10.5°). Together, these results demonstrate that perceptual straightening arises from a cascaded computation that involves at least the early and mid-level processing stages.