Leeds Butterfly Dataset

This dataset contains images and textual descriptions for ten categories (species) of butterflies. More specifically, it contains:

• Images for ten butterfly categories
• Segmentation masks for each image
• Textual descriptions for each butterfly category

The image dataset comprises 832 images in total, with the distribution ranging from 55 to 100 images per category. Images were collected from Google Images by querying with the scientific (Latin) name of the species, for example "Danaus plexippus", and manually filtered for those depicting the butterfly of interest. 

The textual descriptions for each butterfly category were obtained from the eNature online nature guide back in 2008 (website no longer available).

Please refer to our paper for a more detailed description of the dataset:

Josiah Wang, Katja Markert, Mark Everingham (2009). Learning Models for Object Recognition from Natural Language Descriptions. In Proceedings of the 20th British Machine Vision Conference (BMVC2009), September 2009. Also see the video recording of the oral presentation.