1M Unlabeled Distractor Sets

With over 1 milllion unlabeled faces, our distractor set is utilized in challenge 1 to test state-of-the-art algorithms at the million scale.

MF2: 672K Labeled Identities

With 672K identities and 4.7M photos, our training set is the most broad open dataset available. We provide challenge 2 to train on this dataset. We also provide a disjoint 1M distractor set from our MF2 training set to be used in challenge 2.