Herbarium specimens scans (GLM) and associated label data used for zero-shot handwriting recognition of historical herbarium labels

The dataset contains 200 images of herbarium specimens from the Herbarium Senckenbergianum Görlitz (GLM) and the associated label data used by Körschens et al. for their study "Large Language Vision Models for Zero-Shot Handwriting Recognition of Historical Herbarium Labels". One half of the specimen images features handwriten labels, the other half printed labels. All specimens are part of the virtual herbarium of the flora of Germany (https://virtherbard.senckenberg.de). The csv-arquives with the associated label data are an corrected version of the database entries for the label data from the virtual herbarium. The original database entries are not always completely identical to the information on the labels, as additional information (e.g. current species names, full first names instead of initials, information on the natural region) had been added during the digitisation of the data. For the study the database entries were therefore partly adjusted back to the information visible on the label. Date and name information was not corrected back.