UMDFaces: An annotated face dataset for training deep networks

By Ankan Bansal, Anirudh Nanduri, Carlos Domingo Castillo, Rajeev Ranjan, Ramalingam Chellappa. 2016

232 citations recorded by the source corpus.

Recent progress in face detection (including keypoint detection), and recognition is mainly being driven by (i) deeper convolutional neural network architectures, and (ii) larger datasets. However, most of the large datasets are maintained by private companies and are not publicly available. The academic computer vision community needs larger and more varied datasets to make further progress. In this paper, we introduce a new face dataset, called UMDFaces, which has 367,888 annotated faces of 8,277 subjects. We also introduce a new face recognition evaluation protocol which will help advance the state-of-the-art in this area. We discuss how a large dataset can be collected and annotated using human annotators and deep networks. We provide human curated bounding boxes for faces. We also provide estimated pose (roll, pitch and yaw), locations of twenty-one key-points and gender information generated by a pre-trained neural network. In addition, the quality of keypoint annotations has been verified by humans for about 115,000 images. Finally, we compare the quality of the dataset with other publicly available face datasets at similar scales.

Subfields: Computer Vision and Pattern Recognition, 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 Signal Processing

Citations in the corpus

Diversity in Faces (2019, methodology): “Several data sets such as IJBC [53], UMD [26] and VGGFace [22, 23] provide a larger set of face images with a wider range of pose and lighting variations.”
VGGFace2: A Dataset for Recognising Faces across Pose and Age (2017, background): “Unlike the above datasets which are geared towards imagebased face recognition, the Youtube Face (YTF) [23] and UMDFaces-Videos [4] datasets aim to recognise faces in unconstrained videos.”
Beyond Face Rotation: Global and Local Perception GAN for Photorealistic and Identity Preserving Frontal View Synthesis (2017, methodology): “The CNN for gender classiﬁcation is of the same structure as the encoder G θ gE and is trained on batch 1 of the UMD [3] dataset.”
WebFace260M: A Benchmark Unveiling the Power of Million-Scale Deep Face Recognition (2021, background): “a variety of loss functions [59, 47, 57, 54, 73, 14, 45, 67, 33, 32, 68, 66, 12, 27], and growing face benchmarks [26, 37, 49, 88, 63, 29, 30, 74, 36, 84, 41, 8, 7, 38, 21, 64].”
How Computers See Gender (2019, unknown): “This is generally done using training data comprised of large sets of images, which are qualitatively labeled by humans for specific characteristics (e.g. [9]).”
Global-Local GCN: Large-Scale Label Noise Cleansing for Face Recognition (2020, background): “In recent years, the face recognition community witnesses a boost in datasets, from the ﬁrst widely used deep training set CASIA-WebFace (CA-SIA) [54] to MS-Celeb-1M (MS1M) [21], UMDFace [8] and VGGFace2 [12].”
Deep Face Recognition: A Survey (2018, unknown): “ 2019 10.8K 800 18 OC) 87%@ 10 [213]; 4 AR; 65%@ 10 4 [84] bias [181] 2017 11652 3968 2/2.9/3 OC) [37] (VI-A.1) [182] 2016 7,000 14 OC) [135]; [97] (VI-A.1) [183] 2017 13K 5K 2.3 OC) [20] ﬁne-grained [184] 2016 367,920 8,501 43.3 69.30%@ 10 2 [22] pose [169] 2011 3,425 1,595 48/181.3/6,070 Acc [185]”
Low-Resolution Face Recognition in the Wild via Selective Knowledge Distillation (2018, methodology): “The UMDFaces dataset [59] contains 367 , 888 images with annotations from 8 , 419 subjects, which is obtained by crawling public images on the Internet.”
IARPA Janus Benchmark - C: Face Dataset and Protocol (2018, methodology): “UMDFaces also has no clear authority for redistribution.”
ArcFace: Additive Angular Margin Loss for Deep Face Recognition (2018, methodology): “tributes. The ﬁrst attribute is the training data employed to train the model. The identity number of public available training data, such as VGG-Face [31], VGG2-Face [7], CAISAWebFace [48], UMDFaces [6], MS-Celeb-1M [11], and MegaFace [21], ranges from several thousand to half million. Although MS-”
Poisoning Web-Scale Training Datasets is Practical (2023, background): “Critically, the maintainer does not provide a snapshot of the data associated with urli, due either to untenable storage costs [3,4,17,31], privacy concerns [13, 72], or copyright limitations [16].”
A Light CNN for Deep Face Representation With Noisy Labels (2015, unknown): “the scale of the training dataset for CNN has been consistently increasing. Some large-scale face datasets have been released such as CASIA-WebFace [6], CelebFaces+ [4], VGG face dataset [7], UMDFace [9], [10], MS-Celeb-1M [11] and VGGFace2 dataset [12]. However, these large-scale datasets often con”
MFC Datasets: Large-Scale Benchmark Datasets for Media Forensic Challenge Evaluation (2019, background): “UMDFaces (2016) [37] has 367,888 annotated faces of 8,277 subjects.”