Food – 101 Dataset is an image dataset containing 101 food categories. It is mainly used for image classification. There are 101,000 images in total, with 250 test images and 750 training images for each category. The training images have not been cleaned, but all images have been resized with a maximum side length of 512 pixels.

This dataset was released by Stanford University in 2014, and the related paper is "Food-101 — Mining Discriminative Components with Random Forests".