The MegaFace Benchmark: 1 Million Faces for Recognition at Scale The MegaFace benchmark evaluates face recognition performance with increasing numbers of gallery distractors, from 10 to 1 million. It uses two probe sets: FaceScrub (celebrities) and FGNET (photos with age variation). The benchmark assesses identification and verification performance, and evaluates how algorithms perform with varying numbers of distractors, pose, and age. The dataset includes 1 million photos of over 690,000 unique individuals, collected from Flickr. The challenge tests algorithms on this large-scale dataset, revealing that algorithms trained on larger datasets generally perform better. Age-invariant recognition remains challenging, and pose variation significantly affects performance. The benchmark also highlights the importance of training data size and the need for large-scale testing to evaluate face recognition algorithms effectively. The MegaFace dataset and challenge are publicly available for further research and experimentation.The MegaFace Benchmark: 1 Million Faces for Recognition at Scale The MegaFace benchmark evaluates face recognition performance with increasing numbers of gallery distractors, from 10 to 1 million. It uses two probe sets: FaceScrub (celebrities) and FGNET (photos with age variation). The benchmark assesses identification and verification performance, and evaluates how algorithms perform with varying numbers of distractors, pose, and age. The dataset includes 1 million photos of over 690,000 unique individuals, collected from Flickr. The challenge tests algorithms on this large-scale dataset, revealing that algorithms trained on larger datasets generally perform better. Age-invariant recognition remains challenging, and pose variation significantly affects performance. The benchmark also highlights the importance of training data size and the need for large-scale testing to evaluate face recognition algorithms effectively. The MegaFace dataset and challenge are publicly available for further research and experimentation.