Proceedings of the Annual Meeting of the Cognitive Science Society

While modern computational models accurately predict free viewing scanpaths via learned deep neural networks, the in- vestigation of the brain's implementation of the modeled be- havior has been thus far limited to primarily bottom-up vi- sual features. One of the key question in the investigation in the early visual system's implementation of bottom-up visual saliency is how (and whether) it dynamically adapts its sensi- tivity to features of different spatial scales. This paper provides a simple test of whether identical stimuli presented at different spatial scales produce different gaze behavior. We asked sub- jects (n=12) to freely view video stimuli twice each session over two sessions In one session (intervention) the visual scale changed from large (25 degrees of visual angle) to small (10 degrees) between viewings. In the other session (control) the video size was the same. Gaze was more strongly correlated between viewings of the same video size (r=0.265) than differ- ent sizes (r=0.231), independent of whether there was a long (> 24 hours) or short (< 10 min) delay between viewings, im- plying that memory effects are not a strong factor. Although low, these within-subject correlations are higher than the cor- relation of gaze between different subjects viewing identical videos (r=0.195).