FLAIR: Federated Learning Annotated Images is an image dataset released by Apple in 2022 for research in federated learning and privacy-preserving machine learning. The related research paper is "FLAIR: Federated Learning Annotated Image Repository", designed to provide a benchmark environment with real-world data heterogeneity for federated learning algorithms.

The dataset contains approximately 430,000 images from 51,000 Flickr users, aiming to reflect the challenges faced by federated learning in real-world applications. The images are human-annotated, featuring over 1,600 fine-grained labels and 17 coarse-grained categories, supporting machine learning tasks of varying complexity.

Dataset Composition

• user_id: The Flickr user's NSID, used to construct user-based federated learning partitions.
• image_id: The Flickr image ID.
• labels: A list of coarse-grained labels, mapped from fine-grained labels.
• fine_grained_labels: A list of fine-grained labels, representing the subjects in the image.
• partition: Data split markers, divided into train, val, and test.

Citation

@article{song2022flair,
  title={FLAIR: Federated Learning Annotated Image Repository},
  author={Song, Congzheng and Granqvist, Filip and Talwar, Kunal},
  journal={Advances in Neural Information Processing Systems},
  volume={35},
  pages={37792--37805},
  year={2022}
}