The iNaturalist 2017 dataset contains 675,170 training and validation images covering 5,089 natural categories, which belong to 13 supercategories, including Plantae, Insecta, Aves, Mammalia, etc.

The dataset is highly imbalanced, with the number of images per class varying greatly. For example, the largest super-class “Plantae” contains 19,613 images from 2,101 classes, while the smallest super-class “Protozoa” has only 381 images from 4 classes.