Annotated Image Dataset for Label Detection in Natural History Collections (ELIE Module)

This dataset supports the training and evaluation of the label detection module of the ELIE pipeline, designed for object detection in multi-label image datasets. The dataset consists of annotated images in JPEG and Pascal VOC XML format, split into training (2224 images), validation (556 images), and test (278 images) subsets. The data were derived from four collections, LEPPHIL, Bees Bytes, AntWeb, and Picturae_MfN, each annotated with the “label” class. This dataset facilitates reproducibility and benchmarking of label detection performance across diverse digitized natural history image sources.