pith. sign in

arxiv: 2606.31664 · v1 · pith:QYP2GCG3new · submitted 2026-06-30 · 💻 cs.CV · cs.AI

Sparsity-Inducing Divergence Losses for Biometric Verification

Dimitrios Koutsianos , Ladislav Mo\v{s}ner , Yannis Panagakis , Themos Stafylakis This is my paper

Pith reviewed 2026-07-01 06:05 UTC · model grok-4.3

classification 💻 cs.CV cs.AI
keywords alpha-divergenceQ-Marginmargin penaltyface verificationspeaker verificationsparsityfalse acceptance ratebiometric verification
0
0 comments X

The pith

Placing the margin penalty inside the reference measure of alpha-divergence losses improves biometric verification at low false acceptance rates.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces Q-Margin as an alpha-divergence loss that encodes the margin penalty directly into the reference measure of prior probabilities rather than applying geometric penalties to the logits. This formulation is intended to retain the sparsity that appears when alpha exceeds one while producing more discriminative embeddings for face and speaker verification tasks. Standard margin methods such as those in ArcFace and CosFace were built for softmax and do not transfer cleanly to the generalized divergence setting. If the approach holds, it supplies stronger results precisely at the low false-acceptance operating points required for high-security deployments and permits exact training on datasets containing millions of identities.

Core claim

Q-Margin encodes the margin penalty directly into the reference measure of the alpha-divergence, thereby encouraging discriminative embeddings while preserving the sparsity properties of the divergence for alpha greater than one; under identical training recipes this yields competitive or better results than ArcFace and CosFace on the IJB-B and IJB-C benchmarks, with particular gains at low false acceptance rates, and comparable gains on VoxCeleb speaker verification, together with exact memory-efficient training enabled by the extreme sparsity of the resulting posteriors.

What carries the argument

Q-Margin, an alpha-divergence loss that encodes the margin penalty directly into the reference measure (prior probabilities) instead of the logits.

If this is right

  • Q-Margin achieves competitive or superior performance on the IJB-B and IJB-C face verification benchmarks.
  • Q-Margin improves results at low false acceptance rates relative to ArcFace and CosFace trained identically.
  • The extreme sparsity of Q-Margin posteriors permits exact and memory-efficient training on datasets with millions of identities.
  • Similar performance gains appear in speaker verification on VoxCeleb.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same reference-measure construction could be tested on other divergence families to check whether the sparsity-plus-margin benefit generalizes.
  • Because training becomes exact rather than approximate, the method may scale to identity counts far beyond current large-scale biometric sets without additional sampling tricks.
  • The low-FAR emphasis suggests the loss could be combined with threshold-calibration techniques already used in deployed security systems.

Load-bearing premise

Encoding the margin penalty directly into the reference measure will preserve the sparsity properties of the alpha-divergence while producing more discriminative embeddings than standard geometric margins on logits.

What would settle it

A controlled experiment on IJB-C that trains ArcFace, CosFace, and Q-Margin under the identical recipe and finds no improvement or a degradation at low false acceptance rates for Q-Margin.

Figures

Figures reproduced from arXiv: 2606.31664 by Dimitrios Koutsianos, Ladislav Mo\v{s}ner, Themos Stafylakis, Yannis Panagakis.

Figure 1
Figure 1. Figure 1: Challenging pairs from AgeDB-30, CALFW, and CPLFW where our proposed method (Q-Margin, α=1.25, s=32, m=0.2) succeeds, while the ArcFace baseline fails. The top row shows genuine pairs correctly accepted by our model and falsely rejected by the baseline. The bottom row shows impostor pairs correctly rejected by our model and falsely accepted by the baseline. τ ∗ solely on this active set. Since this maximum… view at source ↗
Figure 2
Figure 2. Figure 2: False acceptance vs missed detection probabilities for various losses on IJB-B and IJB-C (left) and VoxCeleb1-H (right), focused on low FARs. For evaluation, we adopted the challenging trial list termed VoxCeleb1-H (hard). It contains 550,894 trials featuring 1,190 speakers. The difficulty of the evaluation set arises from the constraint that trials consist of utterances from speakers of the same gender an… view at source ↗
read the original abstract

Performance in face and speaker verification is largely driven by margin-penalty softmax losses such as CosFace and ArcFace. Recently introduced $\alpha$-divergence loss functions offer a compelling alternative, particularly due to their ability to induce sparse solutions (when $\alpha>1$). However, standard geometric margins are designed for the softmax function and do not naturally extend to this generalized probabilistic framework. In this paper we propose Q-Margin, a novel $\alpha$-divergence loss that introduces a principled probabilistic margin. Unlike conventional methods that apply geometric penalties to the logits (unnormalized log-likelihoods), Q-Margin encodes the margin penalty directly into the reference measure (prior probabilities). This formulation naturally encourages discriminative embeddings while preserving the beneficial sparsity properties of the $\alpha$-divergence. We demonstrate that Q-Margin achieves competitive or superior performance on the challenging IJB-B and IJB-C face verification benchmarks and similarly strong results in speaker verification on VoxCeleb. Crucially, against ArcFace and CosFace baselines trained under an identical recipe, Q-Margin consistently improves at low False Acceptance Rates (FARs), a capability critical for practical high-security applications. Finally, the extreme sparsity of the Q-Margin posteriors enables exact and memory-efficient training, offering a scalable solution for datasets with millions of identities.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 0 minor

Summary. The paper introduces Q-Margin, a novel α-divergence loss for face and speaker verification. It encodes a probabilistic margin directly into the reference measure (prior probabilities) rather than applying geometric penalties to logits, claiming this yields more discriminative embeddings while preserving the sparsity-inducing properties of α-divergence losses (for α>1). Experiments on IJB-B, IJB-C, and VoxCeleb show competitive or superior verification performance versus ArcFace and CosFace baselines trained identically, with consistent gains at low FARs and extreme sparsity enabling memory-efficient training on large identity sets.

Significance. If the central construction is shown to preserve exact sparsity while improving tail behavior, the work would offer a principled route to margin penalties inside generalized divergence losses. This could matter for high-security biometric systems where low-FAR performance and scalable training on millions of identities are both required. The explicit contrast with identical-recipe ArcFace/CosFace baselines is a strength.

major comments (2)
  1. [Abstract] Abstract (and wherever the construction is defined): the assertion that encoding the margin into the reference measure 'naturally encourages discriminative embeddings while preserving the beneficial sparsity properties' is load-bearing for both the low-FAR claim and the memory-efficient training claim, yet the provided text supplies no derivation showing that the modified reference measure leaves the posterior support unchanged for α>1. Without this step, it is impossible to confirm that sparsity is retained exactly rather than approximately.
  2. [Abstract] The empirical claim of 'consistent improvements at low FARs' against identical-recipe baselines rests on the Q-Margin construction simultaneously achieving (a) retained sparsity and (b) better tail discrimination. If the reference-measure adjustment alters the mode or effective divergence, either property could be lost; a concrete check (e.g., support size of the posterior or an ablation isolating the margin term) is needed to secure this link.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments highlighting the need for a formal derivation of sparsity preservation and additional empirical validation. We will revise the manuscript accordingly to strengthen these aspects.

read point-by-point responses

  1. Referee: [Abstract] Abstract (and wherever the construction is defined): the assertion that encoding the margin into the reference measure 'naturally encourages discriminative embeddings while preserving the beneficial sparsity properties' is load-bearing for both the low-FAR claim and the memory-efficient training claim, yet the provided text supplies no derivation showing that the modified reference measure leaves the posterior support unchanged for α>1. Without this step, it is impossible to confirm that sparsity is retained exactly rather than approximately.

    Authors: We agree that an explicit derivation is required to rigorously establish that the modified reference measure leaves the posterior support unchanged for α>1. In the revised manuscript we will add a mathematical derivation in the section defining Q-Margin, proving that the support of the posterior remains identical to the standard α-divergence case and that sparsity is therefore retained exactly. revision: yes

  2. Referee: [Abstract] The empirical claim of 'consistent improvements at low FARs' against identical-recipe baselines rests on the Q-Margin construction simultaneously achieving (a) retained sparsity and (b) better tail discrimination. If the reference-measure adjustment alters the mode or effective divergence, either property could be lost; a concrete check (e.g., support size of the posterior or an ablation isolating the margin term) is needed to secure this link.

    Authors: We will add the requested concrete checks. The revised experiments section will include (i) measurements of posterior support size comparing Q-Margin to the baseline α-divergence loss across training runs and (ii) an ablation that isolates the margin-encoding term while holding all other factors fixed. These results will be reported on the same IJB-B/C and VoxCeleb protocols to directly link the construction to the observed low-FAR gains. revision: yes

Circularity Check

0 steps flagged

No circularity: novel construction with empirical support

full rationale

The paper proposes Q-Margin by directly modifying the reference measure of the α-divergence to incorporate a probabilistic margin, claiming this preserves sparsity for α>1 while improving discriminability. The abstract presents the construction as a new formulation and supports the central claim via direct comparisons to ArcFace/CosFace on IJB-B, IJB-C and VoxCeleb under identical training recipes. No equations, fitted parameters, or self-citations are shown that reduce the claimed performance gains or sparsity preservation to the inputs by construction. The derivation is therefore self-contained as an ansatz whose validity is tested externally rather than forced by prior definitions or self-referential steps.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Only the abstract is available, so no free parameters, axioms, or invented entities can be extracted or audited from the text.

pith-pipeline@v0.9.1-grok · 5779 in / 1060 out tokens · 21250 ms · 2026-07-01T06:05:12.616945+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

41 extracted references · 15 canonical work pages · 1 internal anchor

  1. [2]

    In: 2021 IEEE/CVF International Conference on Computer Vision Work- shops (ICCVW)

    An, X., Zhu, X., Gao, Y., Xiao, Y., Zhao, Y., Feng, Z., Wu, L., Qin, B., Zhang, M., Zhang, D., Fu, Y.: Partial fc: Training 10 million identities on a single ma- chine. In: 2021 IEEE/CVF International Conference on Computer Vision Work- shops (ICCVW). pp. 1445–1449 (2021).https://doi.org/10.1109/ICCVW54120. 2021.00166

    work page doi:10.1109/iccvw54120 2021

  2. [3]

    In: Chaudhuri, K., Sugiyama, M

    Blondel,M.,Martins,A.,Niculae,V.:Learningclassifierswithfenchel-younglosses: Generalized entropies, margins, and algorithms. In: Chaudhuri, K., Sugiyama, M. (eds.) Proceedings of the Twenty-Second International Conference on Artificial Intelligence and Statistics. Proceedings of Machine Learning Research, vol. 89, pp. 606–615. PMLR (16–18 Apr 2019),https:...

    2019

  3. [4]

    In: 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)

    Boutros, F., Damer, N., Kirchbuchner, F., Kuijper, A.: Elasticface: Elastic margin loss for deep face recognition. 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) pp. 1577–1586 (2021).https:// doi.org/10.1109/CVPRW56347.2022.00164

    work page doi:10.1109/cvprw56347.2022.00164 2022

  4. [5]

    In: 2018 13th IEEE International Conference on Automatic Face & Gesture Recognition (FG 2018)

    Cao, Q., Shen, L., Xie, W., Parkhi, O.M., Zisserman, A.: VGGFace2: A Dataset for Recognising Faces across Pose and Age . In: 2018 13th IEEE International Conference on Automatic Face & Gesture Recognition (FG 2018). pp. 67–74. IEEE Computer Society, Los Alamitos, CA, USA (May 2018).https://doi.org/ 10.1109/FG.2018.00020,https://doi.ieeecomputersociety.org...

    work page doi:10.1109/fg.2018.00020 2018

  5. [6]

    In: 2020 25th International Conference on Pattern Recognition (ICPR)

    Chan, C.H., Kittler, J.: Angular sparsemax for face recognition. In: 2020 25th International Conference on Pattern Recognition (ICPR). pp. 10473–10479. IEEE (2021)

    2020

  6. [7]

    In: Chinese conference on biometric recognition

    Chen, S., Liu, Y., Gao, X., Han, Z.: Mobilefacenets: Efficient cnns for accurate real-time face verification on mobile devices. In: Chinese conference on biometric recognition. pp. 428–438. Springer (2018)

    2018

  7. [8]

    In: Proc

    Chung, J.S., Nagrani, A., Zisserman, A.: VoxCeleb2: Deep speaker recognition. In: Proc. Interspeech. pp. 1086–1090 (2018)

    2018

  8. [9]

    In: Globerson, A., Mackey, L., Belgrave, D., Fan, A., Paquet, U., Tomczak, J., Zhang, C

    Dan, J., Liu, Y., Deng, J., Xie, H., Li, S., Sun, B., Luo, S.: Topofr: A closer look at topology alignment on face recognition. In: Globerson, A., Mackey, L., Belgrave, D., Fan, A., Paquet, U., Tomczak, J., Zhang, C. (eds.) Advances in Neural Informa- tion Processing Systems. vol. 37, pp. 37213–37240. Curran Associates, Inc. (2024). https://doi.org/10.522...

    work page doi:10.52202/079017-1174 2024

  9. [10]

    In: 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    Deng, J., Guo, J., Xue, N., Zafeiriou, S.: Arcface: Additive angular margin loss for deep face recognition. In: 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 4685–4694 (2019).https://doi.org/10.1109/ CVPR.2019.00482

    work page arXiv 2019

  10. [11]

    He,K.,Zhang,X.,Ren,S.,Sun,J.:Deepresiduallearningforimagerecognition.In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 770–778 (2016) Sparsity-Inducing Divergence Losses for Biometric Verification 17

    2016

  11. [12]

    Huang, G.B., Mattar, M., Berg, T., Learned-Miller, E.: Labeled faces in the wild: A databaseforstudyingfacerecognitioninunconstrainedenvironments.In:Workshop on faces in’Real-Life’Images: detection, alignment, and recognition (2008)

    2008

  12. [13]

    In: proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Huang, Y., Wang, Y., Tai, Y., Liu, X., Shen, P., Li, S., Li, J., Huang, F.: Curricular- face: adaptive curriculum learning loss for deep face recognition. In: proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 5901– 5910 (2020)

    2020

  13. [14]

    IEEE/ACM Transactions on Audio, Speech, and Language Processing32, 3850– 3866 (2024).https://doi.org/10.1109/TASLP.2024.3444456

    Huh, J., Chung, J.S., Nagrani, A., Brown, A., Jung, J.w., Garcia-Romero, D., Zisserman, A.: The VoxCeleb speaker recognition challenge: A retrospective. IEEE/ACM Transactions on Audio, Speech, and Language Processing32, 3850– 3866 (2024).https://doi.org/10.1109/TASLP.2024.3444456

    work page doi:10.1109/taslp.2024.3444456 2024

  14. [15]

    Ego4d: Around the world in 3, 000 hours of egocentric video

    Kim,M.,Jain,A.K.,Liu,X.:Adaface:Qualityadaptivemarginforfacerecognition. In: 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 18729–18738 (2022).https://doi.org/10.1109/CVPR52688.2022. 01819

    work page doi:10.1109/cvpr52688.2022 2022

  15. [16]

    Kim, M., Liu, F., Jain, A.K., Liu, X.: Cluster and aggregate: Face recognition with large probe set. vol. 35, pp. 36054–36066 (2022)

    2022

  16. [17]

    Advances in neural information processing systems29(2016)

    Li, Y., Turner, R.E.: Rényi divergence variational inference. Advances in neural information processing systems29(2016)

    2016

  17. [18]

    In: 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)

    Liu, W., Wen, Y., Yu, Z., Li, M., Raj, B., Song, L.: Sphereface: Deep hypersphere embedding for face recognition. In: 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). pp. 6738–6746 (2017).https://doi.org/10. 1109/CVPR.2017.713

    2017

  18. [19]

    In: Proceedings of the 33rd International Conference on Inter- national Conference on Machine Learning - Volume 48

    Liu, W., Wen, Y., Yu, Z., Yang, M.: Large-margin softmax loss for convolutional neural networks. In: Proceedings of the 33rd International Conference on Inter- national Conference on Machine Learning - Volume 48. p. 507–516. ICML’16, JMLR.org (2016)

    2016

  19. [20]

    In: Balcan, M.F., Weinberger, K.Q

    Martins, A., Astudillo, R.: From softmax to sparsemax: A sparse model of attention and multi-label classification. In: Balcan, M.F., Weinberger, K.Q. (eds.) Proceed- ings of The 33rd International Conference on Machine Learning. Proceedings of Machine Learning Research, vol. 48, pp. 1614–1623. PMLR, New York, New York, USA (20–22 Jun 2016),https://proceed...

    2016

  20. [21]

    In: 2018 international conference on biometrics (ICB)

    Maze, B., Adams, J., Duncan, J.A., Kalka, N., Miller, T., Otto, C., Jain, A.K., Niggel, W.T., Anderson, J., Cheney, J., et al.: Iarpa janus benchmark-c: Face dataset and protocol. In: 2018 international conference on biometrics (ICB). pp. 158–165. IEEE (2018)

    2018

  21. [22]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    Meng, Q., Zhao, S., Huang, Z., Zhou, F.: Magface: A universal representation for face recognition and quality assessment. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 14225– 14234 (June 2021)

    2021

  22. [23]

    In: proceedings of the IEEE conference on computer vision and pattern recognition workshops

    Moschoglou, S., Papaioannou, A., Sagonas, C., Deng, J., Kotsia, I., Zafeiriou, S.: Agedb: the first manually collected, in-the-wild age database. In: proceedings of the IEEE conference on computer vision and pattern recognition workshops. pp. 51–59 (2017)

    2017

  23. [24]

    In: Proceedings of the 41st International Conference on Machine Learning

    Novello, N., Tonello, A.M.: f-divergence based classification: beyond the use of cross-entropy. In: Proceedings of the 41st International Conference on Machine Learning. ICML’24, JMLR.org (2024)

    2024

  24. [25]

    Le, B., Woo, S.S.: Kappaface: Adaptive additive angular margin loss for deep face recognition

    Oinar, C., M. Le, B., Woo, S.S.: Kappaface: Adaptive additive angular margin loss for deep face recognition. IEEE Access11, 137138–137150 (2023).https: //doi.org/10.1109/ACCESS.2023.3338648 18 D. Koutsianos et al

    work page doi:10.1109/access.2023.3338648 2023

  25. [26]

    In: Korhonen, A., Traum, D., Màrquez, L

    Peters, B., Niculae, V., Martins, A.F.T.: Sparse sequence-to-sequence models. In: Korhonen, A., Traum, D., Màrquez, L. (eds.) Proceedings of the 57th An- nual Meeting of the Association for Computational Linguistics. pp. 1504–1519. Association for Computational Linguistics, Florence, Italy (Jul 2019).https: //doi.org/10.18653/v1/P19-1146,https://aclanthol...

    work page doi:10.18653/v1/p19-1146 2019

  26. [27]

    In: Proceedings of the fourth Berkeley symposium on mathematical statistics and probability, volume 1: con- tributions to the theory of statistics

    Rényi, A.: On measures of entropy and information. In: Proceedings of the fourth Berkeley symposium on mathematical statistics and probability, volume 1: con- tributions to the theory of statistics. vol. 4, pp. 547–562. University of California Press (1961)

    1961

  27. [28]

    In: Forty-second International Conference on Machine Learning (2025),https://openreview.net/forum?id=V1YfPJDliw

    Roulet, V., Liu, T., Vieillard, N., Sander, M.E., Blondel, M.: Loss functions and operators generated by f-divergences. In: Forty-second International Conference on Machine Learning (2025),https://openreview.net/forum?id=V1YfPJDliw

    2025

  28. [29]

    In: Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision

    Saadabadi, M.S.E., Malakshan, S.R., Zafari, A., Mostofa, M., Nasrabadi, N.M.: A quality aware sample-to-sample comparison for face recognition. In: Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision. pp. 6129–6138 (2023)

    2023

  29. [30]

    In: 2016 IEEE winter conference on applications of computer vision (WACV)

    Sengupta, S., Chen, J.C., Castillo, C., Patel, V.M., Chellappa, R., Jacobs, D.W.: Frontal to profile face verification in the wild. In: 2016 IEEE winter conference on applications of computer vision (WACV). pp. 1–9. IEEE (2016)

    2016

  30. [31]

    In: ICASSP 2021 - 2021 IEEE International Confer- ence on Acoustics, Speech and Signal Processing (ICASSP)

    Thienpondt, J., Desplanques, B., Demuynck, K.: The IdLab VoxSRC-20 sub- mission: Large margin fine-tuning and quality-aware score calibration in DNN based speaker verification. In: ICASSP 2021 - 2021 IEEE International Confer- ence on Acoustics, Speech and Signal Processing (ICASSP). pp. 5814–5818 (2021). https://doi.org/10.1109/ICASSP39728.2021.9414600

    work page doi:10.1109/icassp39728.2021.9414600 2021

  31. [32]

    In: The Twelfth International Conference on Learning Representations (2024),https: //openreview.net/forum?id=2cRzmWXK9N

    Wang, C., Jiang, Y., Yang, C., Liu, H., Chen, Y.: Beyond reverse KL: Gen- eralizing direct preference optimization with diverse divergence constraints. In: The Twelfth International Conference on Learning Representations (2024),https: //openreview.net/forum?id=2cRzmWXK9N

    2024

  32. [33]

    In: 2018 IEEE/CVF Conference onComputerVisionandPatternRecognition.pp.5265–5274(2018).https://doi

    Wang, H., Wang, Y., Zhou, Z., Ji, X., Gong, D., Zhou, J., Li, Z., Liu, W.: Cosface: Large margin cosine loss for deep face recognition. In: 2018 IEEE/CVF Conference onComputerVisionandPatternRecognition.pp.5265–5274(2018).https://doi. org/10.1109/CVPR.2018.00552

    work page doi:10.1109/cvpr.2018.00552 2018

  33. [34]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recog- nition

    Wang, K., Wang, S., Zhang, P., Zhou, Z., Zhu, Z., Wang, X., Peng, X., Sun, B., Li, H., You, Y.: An efficient training approach for very large scale face recognition. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recog- nition. pp. 4083–4092 (2022)

    2022

  34. [35]

    Speech Communication162, 103104 (2024)

    Wang, S., Chen, Z., Han, B., Wang, H., Liang, C., Zhang, B., Xiang, X., Ding, W., Rohdin, J., Silnova, A., et al.: Advancing speaker embedding learning: Wespeaker toolkit for research and production. Speech Communication162, 103104 (2024)

    2024

  35. [36]

    In: proceedings of the IEEE conference on computer vision and pattern recognition workshops

    Whitelam, C., Taborsky, E., Blanton, A., Maze, B., Adams, J., Miller, T., Kalka, N., Jain, A.K., Duncan, J.A., Allen, K., et al.: Iarpa janus benchmark-b face dataset. In: proceedings of the IEEE conference on computer vision and pattern recognition workshops. pp. 90–98 (2017)

    2017

  36. [37]

    In: 2022 IEEE 24th International Workshop on Multimedia Signal Processing (MMSP)

    Wuerkaixi, A., Yan, K., Zhang, Y., Duan, Z., Zhang, C.: Dyvise: Dynamic vision- guided speaker embedding for audio-visual speaker diarization. In: 2022 IEEE 24th International Workshop on Multimedia Signal Processing (MMSP). pp. 1–6. IEEE (2022)

    2022

  37. [38]

    In: 2019 Asia-Pacific Signal and Information Processing Association Annual Summit and Sparsity-Inducing Divergence Losses for Biometric Verification 19 Conference (APSIPA ASC)

    Xiang, X., Wang, S., Huang, H., Qian, Y., Yu, K.: Margin matters: Towards more discriminative deep neural network embeddings for speaker recognition. In: 2019 Asia-Pacific Signal and Information Processing Association Annual Summit and Sparsity-Inducing Divergence Losses for Biometric Verification 19 Conference (APSIPA ASC). pp. 1652–1656 (2019).https://d...

    work page arXiv 2019

  38. [39]

    Beijing University of Posts and Telecommu- nications, Tech

    Zheng, T., Deng, W.: Cross-pose lfw: A database for studying cross-pose face recog- nition in unconstrained environments. Beijing University of Posts and Telecommu- nications, Tech. Rep5(7), 5 (2018)

    2018

  39. [40]

    Cross-Age LFW: A Database for Studying Cross-Age Face Recognition in Unconstrained Environments

    Zheng, T., Deng, W., Hu, J.: Cross-age lfw: A database for studying cross-age face recognition in unconstrained environments. arXiv preprint arXiv:1708.08197 (2017)

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  40. [41]

    In: 2023 IEEE/CVF International Conference on Computer Vision (ICCV)

    Zhou, J., Jia, X., Li, Q., Shen, L., Duan, J.: Uniface: Unified cross-entropy loss for deep face recognition. In: 2023 IEEE/CVF International Conference on Computer Vision (ICCV). pp. 20673–20682 (2023).https://doi.org/10.1109/ICCV51070. 2023.01895

    work page doi:10.1109/iccv51070 2023

  41. [42]

    IEEE Transactions on Pattern Analysis and Machine Intelligence45(2), 2627–2644 (2023).https://doi.org/10.1109/TPAMI.2022.3169734

    Zhu, Z., Huang, G., Deng, J., Ye, Y., Huang, J., Chen, X., Zhu, J., Yang, T., Du, D., Lu, J., Zhou, J.: Webface260m: A benchmark for million-scale deep face recog- nition. IEEE Transactions on Pattern Analysis and Machine Intelligence45(2), 2627–2644 (2023).https://doi.org/10.1109/TPAMI.2022.3169734

    work page doi:10.1109/tpami.2022.3169734 2023