Synthetic Images for Medieval Musical Iconography

A dataset of synthetic images representing pages of medieval manuscripts and annotated with bounding boxes for text, instruments, and performers. It consists of three version: one including only instruments, one including only performers, and a mixed one including both. Each version comprises a total of 10k images, divided into train (7k images), val (1.5k images), and test (1.5k images) subsets. The dataset presents a balance in the represented labels, with original annotation extracted from the Musiconis dataset.