Digital Collectomics Example SIP Jena applied ML

This dataset contains a selection of 100 herbarium scans (low-resolution) from the Herbarium Haussknecht, provided by Senckenberg Institute for Plant Form and Function (SIP), Jena. The images have been processed with the machine learning tool (convolutional neural network) for plant organ detection by Younis et al. 2020, see doi.org:10.48550/arXiv.2007.13106 The dataset consists of two files according to the RO-Crate specification (https://www.researchobject.org/ro-crate/1.1/).

derived-images-and-annotations-lowres.zip: A full RO-Crate which contains both the detected plant organ annotations in a machine-readable form as part of the ro-crate-metadata.json and low-res images of the herbarium scans themselves where the detected bounding boxes are visualized
annotations-highres.json: A pure RO-Crate metadata JSON-LD file which contains the detected annotations in reference to the original, high-resolution scans (which are referenced via their web URL)

The annotations consist of the following 6 different classes (mapped to terms from controlled vocabulary):

leaf -> http://purl.obolibrary.org/obo/FLOPO_0000004
flower -> http://purl.obolibrary.org/obo/FLOPO_0000122
fruit -> http://purl.obolibrary.org/obo/FLOPO_0000030
seed -> http://purl.obolibrary.org/obo/FLOPO_0000074
stem -> http://purl.obolibrary.org/obo/FLOPO_0000001
root -> http://purl.obolibrary.org/obo/FLOPO_0000039