The Speed Limit of Visual Perception: Bidirectional influence of image memorability and processing speed on perceived duration and memory recall

Abstract

Visual stimuli are known to vary in their perceived duration, with some stimuli engendering so-called "time dilation" and others "time compression" effects. Previous theories have suggested these effects rely on the level of attention devoted to stimuli, magnitude of the stimulus dimension, or intensity of the population neural response, yet cannot account for the full range of experimental effects. Recently, we demonstrated that perceived time is affected by the image properties of scene clutter, size, and memorability (Ma, et al. 2024), with the former compressing and latter two dilating duration. Further, perceived duration also predicted recall of images 24h later, on top of memorability. To explain the memorability effect, we found that a recurrent convolutional neural network (rCNN) could recapitulate the time dilation effect by indexing the rate of entropy collapse, or "speed", across successive timesteps, with more memorable stimuli associated with faster speeds. Here, we replicate and extend these findings via three experiments (n=20ea.) where subjects performed a sub-second temporal bisection task using memorability stimuli with increasing memorability, but a constant speed (exp 1), increasing speed, but constant memorability (exp 2), or increasing in both (exp 3), each followed by a surprise memory test 24hr later. We found that either increasing memorability or speed alone led to time dilation effects, with faster/slower speeds shifting memory recall by [~]8.4% in either direction. However, when both metrics increased, memorability dilated time while speed compressed it, while still improving recall overall. These findings can be explained by a model wherein the visual system is tuned to a preferred speed for processing stimuli that scales with the magnitude of visual response, such that stimuli closer to this speed are dilated in time. Overall, these findings provide a new lens for interpreting time dilation/compression effects and how visual stimuli are prioritized at temporal scales. Author SummaryRecent work has demonstrated a link between perceived duration and memory recall, with memorable images dilating time, and perceived duration in turn affecting memory. The link between these two can be indexed by the speed at which images are processed in recurrent convolutional neural networks, yet the limits of this link remain untested. Here, we tested three groups on a timing task using three sets of stimuli selected according to both their memorability and speed. We find that when one factor is held constant, both memorability and speed still dilate perceived duration, with speed alone accounting for a shift in memory recall of [~]17%. However, when speeds are raised to their highest limits, perceived duration is instead compressed. An inverted-U shape rule between time dilation and compression over speed that is moderated by memorability explains this effect and suggests the visual system shifts between these modes to both gather more information and conserve energy, respectively.