Multi-Attribute Food (MAFood) is a dataset of food images comprising 3 different tasks: a) dish, b) cuisine and c) categories (food groups). It consists of 21.175 images, distributed as 72.5% for training, 12.5% for validation and 15% for test. Both dish and cuisine can take only one value per image, while categories have multi-label annotations.
In order to choose the images to include, we selected the top 11 most popular cuisines in the world according to Google Trends (www.google.com/trends) (see Figure). For each cuisine, 11 traditional dishes were chosen. In total, the dataset consists of 121 dishes, where each one belongs to at least one of the following 10 food categories: Bread, Egg, Fried food, Meat, Noodles/Pasta, Rice, Seafood, Soup, Dumpling and Vegetable.
Citation
If you use this dataset for any purpose, please, do not forget to cite the following paper:
E. Aguilar, M. Bolaños, P. Radeva, Regularized Uncertainty-based Multi-Task LearningModel for Food Analysis, J. Vis. Commun. Image R. (2019), doi: https://doi.org/10.1016/j.jvcir.2019.03.011