The SemRel database

The SemRel database (zipped file) is a set of standardised images that have been rated for semantic relatedness. This image set specifically targets experiments investigating the influence of top-down expectations on visual recognition, through visual priming, however, the images can also be used for other types of studies involving some form of semantic relatedness between images.

The majority of images included in the dataset are forked from from the Bank of Standardised stimulis (BOSS; Brodeur, Guérard & Bouras, 2014), which is a dataset consisting of 930 standardised images of objects that can be used for free as stimuli for cognitive research. Additional stimuli were selected by browsing google images for freely available images. These images were also controlled for visual complexity, familiarity and size by three independent raters. Finally, visual scenes and backgrounds were selected from the Places database (Zhou, Lapedriza, Khosla, Oliva & Torralba, 2017). This database provided a wide array of visual scenes from different contextual settings which were matched to the objects by the researcher.

Rating for semantic relatedness were obtained in an online study using Limesurvey software. A total of 160 images, distributed across 16 predetermined contextual conditions were rated by 81 participants. The full set of contexts consists of the following: Bathroom, Beach, Bear, Bar/Pub, Crime, Christmas, Kitchen, Music, War, Office, Safari, Snow, Space, Sports, Vehicles, Desert. Semantic relatedness scores were obtained using a 6-point likert scale, ranging from (1) unrelated to (6) related. The exact relatedness scores can be found in the file Relatedness.xlsx