The dataset contains about 28K medium-quality animal images belonging to 10 categories: dog, cat, horse, spider, butterfly, chicken, sheep, cow, squirrel, elephant.

Researchers use it to test different image recognition networks: from homemade CNNs (accuracy of about 80%) to Google Inception (98%). It can simulate a smart gallery for researchers (such as biologists).

All images are from Google Images and have been manually checked. There is some false data to simulate real-world conditions (e.g. images taken by users of your app).

The main directory is divided into multiple folders, one for each category. The number of images per category varies from 2K to 5K units.