Huiqin Chen¹ (), Mei Yang¹, Gaeun Son¹, Dirk Bernhardt-Walther¹; ¹University of Toronto

In dynamically changing environments, how does our visual system make quick perceptual decisions based on information that is actively changing? The hysteresis effect suggests that when resolving perceptual ambiguity, people tend to stick with their current interpretation of sensory information and find it difficult to change their perception until there is a noticeable and significant change in input. We here demonstrate this effect for dynamically changing scenes. We further explored how prior exposure to target scene would affect the hysteresis effect. We used indoor scene images generated with a Generative Adversarial Network to create smooth yet realistic transitions between scene categories. Participants were asked to report when they perceived a shift in category during these transitions. Before each trial, participants were provided with information about the scene category at the end of the transition, either in the form of words, images, or both. Each transition was repeated in both directions (A to B and B to A), and the differences in responses between the two opposite directions were analyzed. Our findings indicated that exposure to words or images had no effect on perceptual hysteresis. Even when participants had knowledge of the target category, their perception of the image categories was still biased towards the initial category. This result suggests that neither semantic knowledge nor visual representation of the future can influence the hysteresis effect. The top-down knowledge of the future direction does not impact or eliminate the conservatism of the visual system. This has important implications, indicating that hysteresis is an innate characteristic of the visual system and is not easily influenced by higher-level control. Thus, perceptual conservatism is likely instrumental for the apparent stability of visual perception in most real-world settings.

Acknowledgements: This work was supported by an NSERC Discovery Grant (RGPIN-2020-04097) to DBW.