Abstract
Knowing where to go and how to get there requires computing spatial directions in visual scenes. Despite high variation in detail across real-world environments, the visual system is quite capable at efficiently and automatically extracting spatial directions from visual scenes (Bonner & Epstein, 2017). People also comprehend spatial directions across distinct formats (words, maps, visual scenes), which require distinct visual computations, yet result in highly similar representations for action (Weisberg, Marchette, & Chatterjee, 2018). Here, in a pre-registered experiment, we collected original behavioral data on 500 images from the Scenes dataset used in the BOLD5000 study (Chang et al., 2019). We recruited 94 subjects to perform one of two behavioral tasks on a normed subset of 100 images each. In a categorical task, subjects indicated which spatial directions were presented by clicking boxes with named spatial directions (e.g., 'left', 'sharp right', 'ahead'). In an analog task, subjects used their mouse to draw up to 7 spatial directions, beginning from the bottom center. After removing subjects who failed an attention check, we calculated the correlation between images for the two behavioral tasks by averaging all participant responses per image (i.e., we calculated the representational dissimilarity matrix, RDM, per task). Comparing behavioral tasks, we found a significant correlation across images (r(500) = .28, p < .001). Exploiting the fMRI dataset collected in the BOLD5000 experiment, we calculated the correlation between each behavioral experiment RDM and the RDM for the BOLD signal in visual scene regions (early visual cortex, OPA, PPA, and RSC). In a preregistered analysis, we observed that correlations between the analog task and visual regions were higher than correlations for the categorical task and visual regions, which did not reach significance. In sum, these findings are a pre-registered replication and suggest a novel dissociation between behavioral representations of visual scenes.