Harnessing large-scale herbarium image datasets through representation learning (application images)

Herbarium specimen images used in the manuscript Harnessing large-scale herbarium image datasets through representation learning for application tasks 2 and 3. These tasks were formulated to test the applicability of representations learned by three different neural networks trained on the Half-Earth Challenge dataset.

All images are for specimens held by Royal Botanic Gardens, Kew. The images from Kew's HerbCat service, using URLs collated by iDigBio. We first search iDigBio for the specimens we wanted, then downloaded the associated occurrence and image metadata files. We then used the URLs in the image metadata files to download the specimen images. The code used to download these images can be found in the GitHub repository for the paper.

We searched iDigBio using these parameters:

[x] Must have media
Genus: Dendrobium
Institution Code: K
Basis of Record: PreservedSpecimen

© All images copyright of the Board of Trustees of the Royal Botanic Gardens, Kew.