Project link: http://megaface.cs.washington.edu/

MegaFace provides a large scale dataset for face recognition. The MegaFace dataset is divided into three parts:

• Training set
• Test set
• Distractors

The Training set contains 4.7 million photos of over 672,057 unique identities. The test set contains the images from the FaceScrub and FGNet dataset. The distractors contain one million photos of 690,572 unique users. Currently, there are two challenges in the MegaFace website. In challenge 1, you can train using any dataset and test your method with the one million distractors. Your method needs to discriminate between a set of known people while classifying the distractors as unknown people. In challenge 2, you will train using the training set with 672K unique identities and test with 1 million distractors. MegaFace is currently the largest dataset for face recognition at the time of writing.