IARPA Janus Benchmark-B Face Dataset

By Cameron Whitelam, Emma Taborsky, Austin Blanton, Brianna Maze, Jocelyn C. Adams, Jocelyn C. Adams, T. Miller, Nathan D. Kalka, Anil K. Jain, J. A. Duncan and 4 more. 2017

482 citations recorded by the source corpus.

Despite the importance of rigorous testing data for evaluating face recognition algorithms, all major publicly available faces-in-the-wild datasets are constrained by the use of a commodity face detector, which limits, among other conditions, pose, occlusion, expression, and illumination variations. In 2015, the NIST IJB-A dataset, which consists of 500 subjects, was released to mitigate these constraints. However, the relatively low number of impostor and genuine matches per split in the IJB-A protocol limits the evaluation of an algorithm at operationally relevant assessment points. This paper builds upon IJB-A and introduces the IARPA Janus Benchmark-B (NIST IJB-B) dataset, a superset of IJB-A. IJB-B consists of 1,845 subjects with human-labeled ground truth face bounding boxes, eye/nose locations, and covariate metadata such as occlusion, facial hair, and skintone for 21,798 still images and 55,026 frames from 7,011 videos. IJB-B was also designed to have a more uniform geographic distribution of subjects across the globe than that of IJB-A. Test protocols for IJB-B represent operational use cases including access point identification, forensic quality media searches, surveillance video searches, and clustering. Finally, all images and videos in IJB-B are published under a Creative Commons distribution license and, therefore, can be freely distributed among the research community.

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

Designing Disaggregated Evaluations of AI Systems: Choices, Considerations, and Tradeoffs (2021, methodology): “Reusing an existing dataset, such as the IARPA Janus Benchmark A (IJB-A) dataset for evaluating face-detection and face-recognition AI systems [77] or the Adience dataset for evaluating age and gender classifiers [22], is less time consuming and less expensive than creating a new dataset.”
Diversity in Faces (2019, methodology): “The gender annotations used two class labels (male and female) and the age group labeling used seven classes ([0-3],[4-12],[13-19],[20-30],[31-45],[46-60],[61-]), as well as a continuous age value to be consistent with the automatic prediction labels.”
VGGFace2: A Dataset for Recognising Faces across Pose and Age (2017, methodology): “The results of [22] are read from the curves reported in the paper.”
Comparator Networks (2018, background): “In § 5, we report the verification performance of the DCN on the challenging IARPA Janus face recognition benchmarks – IJB-B [13] and IJB-C [14].”
Global-Local GCN: Large-Scale Label Noise Cleansing for Face Recognition (2020, methodology): “identities overlapping with LFW [26], FaceScrub [37] and IJB [34, 46].”
MagFace: A Universal Representation for Face Recognition and Quality Assessment (2021, methodology): “Compared to baselines, our MagFace remains the top at all FAR criteria except for FAR=1e-6 on IJB-B as the TAR is very sensitive to the noise when the number of FP is tiny.”
AdaFace: Quality Adaptive Margin for Face Recognition (2022, methodology): “Examples of datasets in this setting are IJB-B [37], IJB-C [22] and IJB-S [14], where most of the images are of low quality, and some do not contain sufficient identity information, even for human examiners.”
CurricularFace: Adaptive Curriculum Learning Loss for Deep Face Recognition (2020, methodology): “We extensively test our method on several popular benchmarks, including LFW [9], CFP-FP [24], CPLFW [41], AgeDB [20], CALFW [40], IJB-B [33], IJB-C [19], and MegaFace [11].”
Sub-center ArcFace: Boosting Face Recognition by Large-Scale Noisy Web Faces (2020, methodology): “Our testing datasets consist of IJB-B [35], IJB-C [17], MegaFace [13], and Face Recognition Vendor Test (FRVT).”
WebFace260M: A Benchmark Unveiling the Power of Million-Scale Deep Face Recognition (2021, background): “Popular evaluations for face recognition including LFW families [26, 88, 63], CFP [49], AgeDB [37], RFW [70] MegaFace [29], IJB families [30, 74, 36] mainly target the pursuit of the accuracy, which have been almost saturated recently.”
Toward More Realistic Face Recognition Evaluation Protocols for the YouTube Faces Database (2018, background): “Several databases are available for the evaluation of unconstrained face recognition in videos [2, 6, 8, 9, 14, 18, 19].”
Deep Face Recognition: A Survey (2018, background): “oposed to explore this issue [93]. Remaining challenges deﬁned by non-saturated benchmark datasets. Three current major datasets, namely, MegaFace [44], [164] , MS-Celeb-1M [45] and IJB-A/B/C [41], [42], [43], are corresponding to largescale FR with a very large number of candidates, low/oneshot FR”
A Light CNN for Deep Face Representation With Noisy Labels (2015, background): “The third type is the very challenging large-scale databases, including the MegaFace [43], IJB-A [44] and IJB-B [45] datasets.”
SphereFace Revived: Unifying Hyperspherical Face Recognition (2021, unknown): “8M train LFW [77] 5,749 13,233 validation AgeDB-30 [78] 568 16,488 validation CALFW [79] 5,749 11,652 validation CPLFW [80] 5,749 12,174 validation CFP [81] 500 7,000 validation VGGFace2_test [75] 500 173k validation IJB-B [82] 1,845 76.”
ElasticFace: Elastic Margin Loss for Deep Face Recognition (2021, methodology): “The IARPA Janus Benchmark-B ((IJB-B) dataset contains 21,798 still images and 55,026 frames from 7,011 videos of 1,845 identities [29].”
Past, Present, and Future of Face Recognition: A Review (2020, unknown): “4 CFP [35] 2016 7000 500 >14 DMFD [37] 2016 2460 410 IJB-B [40] 2 7 21,798 1845 ≈36.”
ShrinkTeaNet: Million-scale Lightweight Face Recognition via Shrinking Teacher-Student Networks (2019, background): “Similar to the Megaface protocols, ShrinkTeaNet is able to boost the performance of the light-weight backbones significantly and reduce the performance gap to the large-scale backbone to only 0.019 on IJB-B and 0.016 on IJB-C.”
ArcFace: Additive Angular Margin Loss for Deep Face Recognition (2018, methodology): “Results on IJB-B and IJB-C.”