The MegaFace Benchmark: 1 Million Faces for Recognition at Scale

By Ira Kemelmacher-Shlizerman, S. M. Seitz, Daniel Miller, Evan Brossard. 2015

923 citations recorded by the source corpus.

Recent face recognition experiments on a major benchmark (LFW [15]) show stunning performance-a number of algorithms achieve near to perfect score, surpassing human recognition rates. In this paper, we advocate evaluations at the million scale (LFW includes only 13K photos of 5K people). To this end, we have assembled the MegaFace dataset and created the first MegaFace challenge. Our dataset includes One Million photos that capture more than 690K different individuals. The challenge evaluates performance of algorithms with increasing numbers of "distractors" (going from 10 to 1M) in the gallery set. We present both identification and verification performance, evaluate performance with respect to pose and a persons age, and compare as a function of training data size (#photos and #people). We report results of state of the art and baseline algorithms. The MegaFace dataset, baseline code, and evaluation scripts, are all publicly released for further experimentations1.

Subfields: Computer Vision and Pattern Recognition, Evaluation & Benchmarks, Signal Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

In Evaluation & Benchmarks

In Signal Processing

Citations in the corpus

FairFace: Face Attribute Dataset for Balanced Race, Gender, and Age (2019, background): “To date, numerous large scale face image datasets [22, 31, 13, 69, 37, 24, 41, 68, 15, 27, 46, 8, 39] have been proposed and fostered research and development for automated face detection [35, 21], alignment [66, 44], recognition [57, 49], generation [67, 5, 26, 58], modiﬁcation [3, 32, 19], and…”
Diversity in Faces (2019, methodology): “Other large-scale face recognition data sets include MegaFace [13], MS-Celeb [14] and CASIA [25].”
VoxCeleb2: Deep Speaker Recognition (2018, background): “Such a mapping has been learnt effectively for face images, through the use of deep CNN architectures [1, 2, 3] trained on large-scale face datasets [4, 5, 6].”
SphereFace: Deep Hypersphere Embedding for Face Recognition (2017, background): “Typically, face recognition can be categorized as face identification and face verification [8, 11].”
VGGFace2: A Dataset for Recognising Faces across Pose and Age (2017, methodology): “MegaFace dataset [12] was released in 2016 to evaluate face recognition methods with up to a million distractors in the gallery image set.”
Large-Scale Long-Tailed Recognition in an Open World (2019, methodology): “To inspect the generalization ability of our approach, the performance is evaluated on the MegaFace benchmark [23], which has no identity overlap with MS1M-ArcFace.”
MS-Celeb-1M: A Dataset and Benchmark for Large-Scale Face Recognition (2016, background): “Currently, the MegaFace in [11] might be one of the most diﬃcult face identi-ﬁcation benchmarks.”
CosFace: Large Margin Cosine Loss for Deep Face Recognition (2018, methodology): “The training data for Megaface Challenge 2 contains 4.7 million faces and 672K identities, which corresponds to the large protocol.”
Turning a Blind Eye: Explicit Removal of Biases and Variation from Deep Neural Network Embeddings (2018, background): “Large publicly available face datasets are often composed of celebrities, such as [6, 7, 8, 9, 10, 11].”
ETH-XGaze: A Large Scale Dataset for Gaze Estimation under Extreme Head Pose and Gaze Variation (2020, background): “Having public benchmark frameworks for evaluation of popular algorithms is common for many computer vision tasks such as image classification [28], face recognition [17], pedestrian detection [8] and hand pose estimation [43].”
Data Uncertainty Learning in Face Recognition (2020, methodology): “2 benchmarks including LFW [18] and MegaFace [20] 5 , and 3 unconstrained benchmarks: CFP [34] 6 , YTF [43] and IJB-C [28], are used to evaluate the performance of DUL cls/rgs following the standard evaluation protocols.”
FairFace: Face Attribute Dataset for Balanced Race, Gender, and Age for Bias Measurement and Mitigation (2021, background): “To date, numerous large scale face image datasets [21, 31, 13, 70, 37, 23, 43, 69, 14, 26, 48, 8, 40] have been proposed and fostered research and development for automated face detection [35, 20], alignment [67, 46], recognition [57, 51], generation [68, 5, 25, 58], modiﬁcation [3, 32, 18], and…”
One-shot Face Recognition by Promoting Underrepresented Classes (2017, background): “The benchmark task in MegaFace [10] focuses on 80 identities for the query set to be recognized, though millions of images provided as distractors.”
UMDFaces: An annotated face dataset for training deep networks (2016, background): “The recently released Megaface challenge [12] might be the most difﬁcult recognition (identiﬁcation) benchmark currently.”
Sub-center ArcFace: Boosting Face Recognition by Large-Scale Noisy Web Faces (2020, methodology): “Results on MegaFace [13].”
ShrinkTeaNet: Million-scale Lightweight Face Recognition via Shrinking Teacher-Student Networks (2019, background): “MegaFace [45] provides a very challenging testing protocol with million-scale of distractors.”
AdaCos: Adaptively Scaling Cosine Logits for Effectively Learning Deep Face Representations (2019, methodology): “We then evaluate the performance of proposed AdaCos on the MegaFace Challenge [16], which is a publicly available identification benchmark, widely used to test the performance of facial recognition algorithms.”