Detecting meaning in RSVP at 13 ms per picture

By Mary Potter, B. Wyble, C. Hagmann, Emily S. McCourt. Attention, Perception, & Psychophysics, 2013

446 citations recorded by the source corpus.

The visual system is exquisitely adapted to the task of extracting conceptual information from visual input with every new eye fixation, three or four times a second. Here we assess the minimum viewing time needed for visual comprehension, using rapid serial visual presentation (RSVP) of a series of six or 12 pictures presented at between 13 and 80 ms per picture, with no interstimulus interval. Participants were to detect a picture specified by a name (e.g., smiling couple) that was given just before or immediately after the sequence. Detection improved with increasing duration and was better when the name was presented before the sequence, but performance was significantly above chance at all durations, whether the target was named before or only after the sequence. The results are consistent with feedforward models, in which an initial wave of neural activity through the ventral stream is sufficient to allow identification of a complex visual stimulus in a single forward pass. Although we discuss other explanations, the results suggest that neither reentrant processing from higher to lower levels nor advance information about the stimulus is necessary for the conscious detection of rapidly presented, complex visual information.

View Paper: https://www.semanticscholar.org/paper/5f6805572d120b7af24a05cda01141a3a64eeb94

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Citations in the corpus

Deep Neural Networks Rival the Representation of Primate IT Cortex for Core Visual Object Recognition (2014, result): “This choice is justified by previous results that indicate human subjects are performant on similar tasks with just 13 ms presentation times [4], that human performance on similar tasks rapidly increases from 14 ms to 56 ms and has diminishing returns between 56 ms and 111 ms [3], that decoding from”
Human and DNN Classification Performance on Images With Quality Distortions (2019, methodology): “The goal of this experiment is to test the ability of the HVS to use 100ms “gist”-level information to recognize distorted images; 100ms was chosen because it has been shown that humans can accurately recognize images at display times between 13 and 111 ms (Keysers et al. 2001; Potter et al. 2014).”
Toward Interconnected Virtual Reality: Opportunities, Challenges, and Enablers (2017, methodology): “visual and auditory information has to be taken in through the camera and sent to the fog/cloud for processing, with digital information sent back to be precisely overlaid onto the real-world environment, and all this has to happen in less time than it takes for humans to start noticing lag (no more”
Seeing and Conceptualizing: Modularity and the Shallow Contents of Perception (2018, background): “Reentrant loops in the visual system are estimated to take at least 50 ms to make a round trip (Macknik and Martinez-Conde 2007; Dicarlo et al. 2012; Potter et al. 2014).”
HYPE: A Benchmark for Human eYe Perceptual Evaluation of Generative Models (2019, background): “e identifying images that require more evaluators (Weld et al., 2015). We also aim to build in faster time exposures under 100ms — ideally down to 13ms, the minimum time exposure of human perception (Potter et al., 2014) — for tasks that require that level of granularity. Doing so requires careful e”
Attacking Automatic Video Analysis Algorithms: A Case Study of Google Cloud Video Intelligence API (2017, background): “Speed of processing in the human visual system largely depends on the contrast of successive images shown [37, 48].”