This GitHub repository contains Python notebooks and data (stimuli) that we used to investigate the visual number sense capabilities of some Foundation Models.
Images generated in the numerosity production task are labeled according to the following format: n1_n2_category_n3.png, where n1 represents the target number, n2 represent the generated number, n3 is just an index for the file, and "category" can be any of the following: apples, butterflies, dots, people.