Existing image classification datasets used in computer vision tend to have an even number of images for each object category. In contrast, the natural world is heavily imbalanced, as some species are more abundant and easier to photograph than others. To encourage further progress in challenging real world conditions we present the iNaturalist Challenge 2017 dataset - an image classification benchmark consisting of 675,000 images with over 5,000 different species of plants and animals. It features many visually similar species, captured in a wide variety of situations, from all over the world. Images were collected with different camera types, have varying image quality, have been verified by multiple citizen scientists, and feature a large class imbalance. We discuss the collection of the dataset and present baseline results for state-of-the-art computer vision classification models. Results show that current non-ensemble based methods achieve only 64% top one classification accuracy, illustrating the difficulty of the dataset. Finally, we report results from a competition that was held with the data.
View on arXiv