pith. machine review for the scientific record. sign in

arxiv: 2605.04769 · v1 · submitted 2026-05-06 · 💻 cs.CV

Recognition: unknown

Lightweight Cross-Spectral Face Recognition via Contrastive Alignment and Distillation

Anjith George , Sebastien Marcel
Authors on Pith no claims yet

Pith reviewed 2026-05-08 17:00 UTC · model grok-4.3

classification 💻 cs.CV
keywords heterogeneous face recognitioncross-spectral matchinglightweight modelscontrastive alignmentknowledge distillationCNN-Transformerthermal-visible recognition
0
0 comments X

The pith

Adapting a hybrid CNN-Transformer model creates a lightweight framework for cross-spectral face recognition that trains on small paired datasets while preserving RGB performance.

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

The paper seeks to make heterogeneous face recognition usable on devices with limited power by creating a compact model that handles images from different sensors, such as thermal and visible light. It starts from a hybrid CNN-Transformer built for ordinary face recognition and modifies it with contrastive alignment plus distillation so that only a modest set of paired cross-modal images is needed for training. The same model continues to work well on standard single-modality face tasks without extra overhead. This combination addresses the practical barrier that heavier HFR methods face when they must run on edge hardware.

Core claim

The central claim is that a hybrid CNN-Transformer architecture originally developed for RGB face recognition can be adapted for heterogeneous face recognition through contrastive alignment and distillation, allowing efficient end-to-end training with only a small amount of paired heterogeneous data while still delivering state-of-the-art or competitive results on both HFR benchmarks and standard RGB face recognition tasks at low computational cost.

What carries the argument

Contrastive alignment and distillation applied to an adapted hybrid CNN-Transformer backbone that aligns features across modalities without enlarging the model or requiring large paired datasets.

If this is right

  • The same model handles both cross-spectral and ordinary RGB face recognition without needing separate architectures.
  • Training remains feasible when only limited paired images exist across sensor types such as NIR-visible or thermal-visible.
  • Computational cost stays low enough for deployment on resource-constrained hardware.
  • Performance on standard face-recognition benchmarks does not degrade after the adaptation.

Where Pith is reading between the lines

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

  • The small paired-data requirement could simplify adaptation to new sensor pairs in domains beyond faces, such as medical or satellite imagery.
  • Keeping the backbone unchanged opens the possibility of swapping in newer lightweight transformers without redesigning the alignment steps.
  • Real-time cross-modal verification on mobile devices becomes more practical if the low-compute property holds across additional modalities.

Load-bearing premise

The hybrid CNN-Transformer backbone originally trained on RGB data can be successfully repurposed for cross-modal matching using only small amounts of paired heterogeneous images while retaining its original accuracy on homogeneous tasks.

What would settle it

An ablation test on a thermal-to-visible benchmark where the contrastive alignment and distillation losses are removed and accuracy falls below current competitive HFR methods without any increase in model size.

Figures

Figures reproduced from arXiv: 2605.04769 by Anjith George, Sebastien Marcel.

Figure 1
Figure 1. Figure 1: Model architecture of xEdgeFace models: The highlighted modules (LN-LayerNorm, ST-Conv. Stem, Stages-S0, S1, S2) are adapted while view at source ↗
Figure 2
Figure 2. Figure 2: Performance evolution using different fractions of the training view at source ↗
Figure 3
Figure 3. Figure 3: Genuine and Impostor Score Distribution Before and After Adap view at source ↗
Figure 4
Figure 4. Figure 4: t-SNE plots of visible and thermal images at different stages of the view at source ↗
Figure 5
Figure 5. Figure 5: Illustration of successful and failure cases in the MCXFace view at source ↗
read the original abstract

Heterogeneous Face Recognition (HFR) aims at matching face images captured across different sensing modalities, such as thermal-to-visible or near-infrared-to-visible, enhancing the usability of face recognition systems in challenging real-world conditions. Although recent HFR methods have achieved significant improvements in performance, many rely on computationally expensive models, making them impractical for deployment on resource-limited edge devices. In this work, we introduce a lightweight yet effective HFR framework by adapting a hybrid CNN-Transformer model originally developed for RGB homogeneous face recognition. Our approach enables efficient end-to-end training with only a small amount of paired heterogeneous data, while still maintaining strong performance on standard RGB face recognition benchmarks. This makes it suitable for both homogeneous and heterogeneous settings. Comprehensive experiments on several challenging HFR and face recognition benchmarks show that our method achieves state-of-the-art or competitive performance while keeping computational requirements low.

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

1 major / 2 minor

Summary. The manuscript introduces a lightweight framework for heterogeneous face recognition (HFR) by adapting a hybrid CNN-Transformer model pretrained on RGB data. It employs contrastive alignment and distillation to support efficient end-to-end training using only a small amount of paired cross-spectral data, while claiming to retain strong performance on both HFR benchmarks (e.g., NIR-VIS) and homogeneous RGB face recognition tasks with low computational overhead.

Significance. If the performance claims hold, the work would provide a practical advance for edge-device deployment of cross-modal face recognition by reducing reliance on large paired heterogeneous datasets. The adaptation of an established RGB architecture via alignment and distillation could bridge homogeneous and heterogeneous settings efficiently. However, the significance is limited by insufficient experimental validation of the data-efficiency premise.

major comments (1)
  1. [Section 4] Section 4 (Experiments): No ablation varies the fraction of paired heterogeneous training data (e.g., 5%, 10%, 20% subsets of CASIA NIR-VIS or similar) while reporting rank-1 accuracy or TAR@FAR curves. This directly undermines the Abstract claim of 'efficient end-to-end training with only a small amount of paired heterogeneous data,' as it is impossible to determine whether the reported SOTA/competitive results require the full paired set or are largely carried by RGB pre-training alone.
minor comments (2)
  1. [Abstract] Abstract: the assertion of 'comprehensive experiments showing SOTA or competitive results' should be supported by explicit mention of datasets, metrics, and at least one quantitative comparison in the abstract itself.
  2. [Section 3] The description of the contrastive alignment and distillation losses in Section 3 would be clearer with explicit equations and hyperparameter values rather than high-level prose.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their constructive feedback on our manuscript. The major comment regarding experimental validation of data efficiency is well-taken, and we address it point-by-point below with plans for revision.

read point-by-point responses

  1. Referee: [Section 4] Section 4 (Experiments): No ablation varies the fraction of paired heterogeneous training data (e.g., 5%, 10%, 20% subsets of CASIA NIR-VIS or similar) while reporting rank-1 accuracy or TAR@FAR curves. This directly undermines the Abstract claim of 'efficient end-to-end training with only a small amount of paired heterogeneous data,' as it is impossible to determine whether the reported SOTA/competitive results require the full paired set or are largely carried by RGB pre-training alone.

    Authors: We agree that an explicit ablation on varying fractions of paired heterogeneous data would strengthen the data-efficiency premise highlighted in the abstract. Our design uses contrastive alignment and distillation to transfer knowledge from the RGB-pretrained hybrid CNN-Transformer backbone, enabling effective adaptation with limited paired samples, but the reported results follow standard HFR protocol by using full paired training sets for direct comparability with prior work. To address this, we will add the requested ablation in the revised Section 4, evaluating performance on 5%, 10%, 20%, and 50% random subsets of the paired training data from CASIA NIR-VIS (and similarly for other benchmarks if space permits), reporting rank-1 accuracy and TAR@FAR curves. These new results will isolate the contribution of our alignment and distillation modules beyond RGB pre-training alone. revision: yes

Circularity Check

0 steps flagged

No circularity detected; derivation relies on independent adaptation of established components

full rationale

The paper adapts a pre-existing hybrid CNN-Transformer architecture (originally for RGB face recognition) via standard contrastive alignment and distillation losses to handle cross-spectral data. No equations, predictions, or first-principles results are shown to reduce by construction to fitted inputs or self-referential definitions. Performance claims rest on experimental benchmarks rather than tautological equivalence, and no load-bearing self-citations or uniqueness theorems imported from the authors' prior work close the derivation loop. The approach is self-contained against external benchmarks and standard ML techniques.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Only the abstract is available; no explicit free parameters, axioms, or invented entities are detailed in the provided text. Standard deep learning assumptions such as the utility of pre-trained models and paired data are implicitly relied upon.

axioms (1)
  • domain assumption Pre-trained RGB face recognition models can be effectively adapted to heterogeneous modalities with limited paired data.
    Abstract states adaptation of an RGB model enables training with small paired heterogeneous data.

pith-pipeline@v0.9.0 · 5442 in / 1193 out tokens · 58399 ms · 2026-05-08T17:00:08.486623+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

88 extracted references · 8 canonical work pages · 2 internal anchors

  1. [1]

    Labeled faces in the wild: A survey,

    E. Learned-Miller, G. B. Huang, A. RoyChowdhury, H. Li, and G. Hua, “Labeled faces in the wild: A survey,”Advances in face detection and facial image analysis, vol. 1, pp. 189–248, 2016

    2016

  2. [2]

    Illumination invariant face recognition using near-infrared images,

    S. Z. Li, R. Chu, S. Liao, and L. Zhang, “Illumination invariant face recognition using near-infrared images,”IEEE Transactions on pattern analysis and machine intelligence, vol. 29, no. 4, pp. 627–639, 2007

    2007

  3. [3]

    A comprehensive evaluation on multi-channel biometric face presentation attack detection,

    A. George, D. Geissbuhler, and S. Marcel, “A comprehensive evaluation on multi-channel biometric face presentation attack detection,”arXiv preprint arXiv:2202.10286, 2022

    work page arXiv 2022

  4. [4]

    Heterogeneous face recognition using kernel prototype similarities,

    B. F. Klare and A. K. Jain, “Heterogeneous face recognition using kernel prototype similarities,”IEEE transactions on pattern analysis and machine intelligence, vol. 35, no. 6, pp. 1410–1422, 2012

    2012

  5. [5]

    Wasserstein CNN: Learning invariant features for Nir-Vis face recognition,

    R. He, X. Wu, Z. Sun, and T. Tan, “Wasserstein CNN: Learning invariant features for Nir-Vis face recognition,”IEEE transactions on pattern analysis and machine intelligence, vol. 41, no. 7, pp. 1761–1773, 2018

    2018

  6. [6]

    Transformers in vision: A survey,

    S. Khan, M. Naseer, M. Hayat, S. W. Zamir, F. S. Khan, and M. Shah, “Transformers in vision: A survey,”ACM computing surveys (CSUR), vol. 54, no. 10s, pp. 1–41, 2022

    2022

  7. [7]

    Edgeface: Efficient face recognition model for edge devices,

    A. George, C. Ecabert, H. O. Shahreza, K. Kotwal, and S. Marcel, “Edgeface: Efficient face recognition model for edge devices,”IEEE Transactions on Biometrics, Behavior, and Identity Science, vol. 6, no. 2, pp. 158–168, 2024

    2024

  8. [8]

    Heterogeneous face recog- nition from local structures of normalized appearance,

    S. Liao, D. Yi, Z. Lei, R. Qin, and S. Z. Li, “Heterogeneous face recog- nition from local structures of normalized appearance,” inInternational Conference on Biometrics. Springer, 2009, pp. 209–218

    2009

  9. [9]

    Matching forensic sketches to mug shot photos,

    B. Klare, Z. Li, and A. K. Jain, “Matching forensic sketches to mug shot photos,”IEEE transactions on pattern analysis and machine intelligence, vol. 33, no. 3, pp. 639–646, 2010

    2010

  10. [10]

    Learning invariant deep represen- tation for Nir-Vis face recognition,

    R. He, X. Wu, Z. Sun, and T. Tan, “Learning invariant deep represen- tation for Nir-Vis face recognition,” inThirty-First AAAI Conference on Artificial Intelligence, 2017

    2017

  11. [11]

    A novel quaternary pattern of local max- imum quotient for heterogeneous face recognition,

    H. Roy and D. Bhattacharjee, “A novel quaternary pattern of local max- imum quotient for heterogeneous face recognition,”Pattern Recognition Letters, vol. 113, pp. 19–28, 2018

    2018

  12. [12]

    Composite components- based face sketch recognition,

    D. Liu, J. Li, N. Wang, C. Peng, and X. Gao, “Composite components- based face sketch recognition,”Neurocomputing, vol. 302, pp. 46–54, 2018

    2018

  13. [13]

    Towards robust facial recognition: Gabor filter-based feature extraction for nir-vis heterogeneous face recognition,

    J. V . de Andrade, A. Freire, G. C. Pereira, C. Millan-Arias, B. Fernandes, C. Bastos-Filho, J. Tortato, L. Da Rocha, and A. M. Maciel, “Towards robust facial recognition: Gabor filter-based feature extraction for nir-vis heterogeneous face recognition,” inProceedings of the 40th ACM/SI- GAPP Symposium on Applied Computing, 2025, pp. 1275–1281

    2025

  14. [14]

    Face matching between near infrared and visible light images,

    D. Yi, R. Liu, R. Chu, Z. Lei, and S. Z. Li, “Face matching between near infrared and visible light images,” inInternational Conference on Biometrics. Springer, 2007, pp. 523–530

    2007

  15. [15]

    Bypassing synthesis: PLS for face recognition with pose, low-resolution and sketch,

    A. Sharma and D. W. Jacobs, “Bypassing synthesis: PLS for face recognition with pose, low-resolution and sketch,” inCVPR 2011. IEEE, 2011, pp. 593–600

    2011

  16. [16]

    Coupled spectral regression for matching heteroge- neous faces,

    Z. Lei and S. Z. Li, “Coupled spectral regression for matching heteroge- neous faces,” in2009 IEEE Conference on Computer Vision and Pattern Recognition. IEEE, 2009, pp. 1123–1128

    2009

  17. [17]

    Heterogeneous face recognition using domain specific units,

    T. de Freitas Pereira, A. Anjos, and S. Marcel, “Heterogeneous face recognition using domain specific units,”IEEE Transactions on Informa- tion Forensics and Security, vol. 14, no. 7, pp. 1803–1816, 2018

    2018

  18. [18]

    Heterogeneous face recognition using domain invariant units,

    A. George and S. Marcel, “Heterogeneous face recognition using domain invariant units,” inICASSP 2024-2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2024, pp. 4780–4784

    2024

  19. [19]

    Coupled attribute learning for heterogeneous face recognition,

    D. Liu, X. Gao, N. Wang, J. Li, and C. Peng, “Coupled attribute learning for heterogeneous face recognition,”IEEE Transactions on Neural Networks and Learning Systems, vol. 31, no. 11, pp. 4699–4712, 2020

    2020

  20. [20]

    Modality-agnostic augmented multi-collaboration representation for semi-supervised heterogenous face recognition,

    D. Liu, W. Yang, C. Peng, N. Wang, R. Hu, and X. Gao, “Modality-agnostic augmented multi-collaboration representation for semi-supervised heterogenous face recognition,” inProceedings of the 31st ACM International Conference on Multimedia, 2023, pp. 4647– 4656

    2023

  21. [21]

    From modalities to styles: Rethinking the domain gap in heterogeneous face recognition,

    A. George and S. Marcel, “From modalities to styles: Rethinking the domain gap in heterogeneous face recognition,”IEEE Transactions on Biometrics, Behavior, and Identity Science, vol. 6, no. 4, pp. 475–485, 2024

    2024

  22. [22]

    Bridging the Gap: Heterogeneous face recognition with condi- tional adaptive instance modulation,

    ——, “Bridging the Gap: Heterogeneous face recognition with condi- tional adaptive instance modulation,” in2023 International Joint Confer- ence on Biometrics (IJCB). IEEE, 2023

    2023

  23. [23]

    Modality agnostic heterogeneous face recognition with switch style modulators,

    ——, “Modality agnostic heterogeneous face recognition with switch style modulators,” in2024 IEEE International Joint Conference on Biometrics (IJCB). IEEE, 2024, pp. 1–10

    2024

  24. [24]

    Face photo-sketch synthesis and recognition,

    X. Wang and X. Tang, “Face photo-sketch synthesis and recognition,” IEEE transactions on pattern analysis and machine intelligence, vol. 31, no. 11, pp. 1955–1967, 2008

    1955

  25. [25]

    A nonlinear approach for face sketch synthesis and recognition,

    Q. Liu, X. Tang, H. Jin, H. Lu, and S. Ma, “A nonlinear approach for face sketch synthesis and recognition,” in2005 IEEE Computer Society conference on computer vision and pattern recognition (CVPR’05), vol. 1. IEEE, 2005, pp. 1005–1010

    2005

  26. [26]

    Unpaired Image- to-Image Translation using Cycle-Consistent Adversarial Networks,

    J.-Y . Zhu, T. Park, P. Isola, and A. A. Efros, “Unpaired Image- to-Image Translation using Cycle-Consistent Adversarial Networks,” arXiv:1703.10593 [cs], Mar. 2017

    work page arXiv 2017

  27. [27]

    Generative adversar- ial network-based synthesis of visible faces from polarimetric thermal faces,

    H. Zhang, V . M. Patel, B. S. Riggan, and S. Hu, “Generative adversar- ial network-based synthesis of visible faces from polarimetric thermal faces,” in2017 IEEE International Joint Conference on Biometrics (IJCB). IEEE, 2017, pp. 100–107

    2017

  28. [28]

    DVG-face: Dual variational generation for heterogeneous face recognition,

    C. Fu, X. Wu, Y . Hu, H. Huang, and R. He, “DVG-face: Dual variational generation for heterogeneous face recognition,”IEEE Transactions on Pattern Analysis and Machine Intelligence, 2021

    2021

  29. [29]

    Heterogeneous face interpretable disentangled representation for joint face recognition and synthesis,

    D. Liu, X. Gao, C. Peng, N. Wang, and J. Li, “Heterogeneous face interpretable disentangled representation for joint face recognition and synthesis,”IEEE transactions on neural networks and learning systems, vol. 33, no. 10, pp. 5611–5625, 2021

    2021

  30. [30]

    Memory-modulated trans- former network for heterogeneous face recognition,

    M. Luo, H. Wu, H. Huang, W. He, and R. He, “Memory-modulated trans- former network for heterogeneous face recognition,”IEEE Transactions on Information Forensics and Security, 2022

    2022

  31. [31]

    Prepended domain trans- former: Heterogeneous face recognition without bells and whistles,

    A. George, A. Mohammadi, and S. Marcel, “Prepended domain trans- former: Heterogeneous face recognition without bells and whistles,” IEEE Transactions on Information Forensics and Security, 2022

    2022

  32. [32]

    Robust cross-domain pseudo- labeling and contrastive learning for unsupervised domain adaptation nir- vis face recognition,

    Y . Yang, W. Hu, H. Lin, and H. Hu, “Robust cross-domain pseudo- labeling and contrastive learning for unsupervised domain adaptation nir- vis face recognition,”IEEE Transactions on Image Processing, vol. 32, pp. 5231–5244, 2023

    2023

  33. [33]

    Pseudo label association and prototype- based invariant learning for semi-supervised nir-vis face recognition,

    W. Hu, Y . Yang, and H. Hu, “Pseudo label association and prototype- based invariant learning for semi-supervised nir-vis face recognition,” IEEE Transactions on Image Processing, vol. 33, pp. 1448–1463, 2024. JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015 11

    2024

  34. [34]

    Unsupervised nir-vis face recognition via homogeneous-to-heterogeneous learning and residual-invariant enhance- ment,

    Y . Yang, W. Hu, and H. Hu, “Unsupervised nir-vis face recognition via homogeneous-to-heterogeneous learning and residual-invariant enhance- ment,”IEEE Transactions on Information Forensics and Security, vol. 19, pp. 2112–2126, 2023

    2023

  35. [35]

    Mobilefacenets: Efficient cnns for accurate real-time face verification on mobile devices,

    S. Chen, Y . Liu, X. Gao, and Z. Han, “Mobilefacenets: Efficient cnns for accurate real-time face verification on mobile devices,” inBiometric Recognition: 13th Chinese Conference, CCBR 2018, Urumqi, China, August 11-12, 2018, Proceedings 13. Springer, 2018, pp. 428–438

    2018

  36. [36]

    MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications

    A. G. Howard, M. Zhu, B. Chen, D. Kalenichenko, W. Wang, T. Weyand, M. Andreetto, and H. Adam, “Mobilenets: Efficient convo- lutional neural networks for mobile vision applications,”arXiv preprint arXiv:1704.04861, 2017

    work page internal anchor Pith review arXiv 2017

  37. [37]

    Mobilenetv2: Inverted residuals and linear bottlenecks,

    M. Sandler, A. Howard, M. Zhu, A. Zhmoginov, and L.-C. Chen, “Mobilenetv2: Inverted residuals and linear bottlenecks,” inProceedings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 4510–4520

    2018

  38. [38]

    Mixfacenets: Extremely efficient face recognition networks,

    F. Boutros, N. Damer, M. Fang, F. Kirchbuchner, and A. Kuijper, “Mixfacenets: Extremely efficient face recognition networks,” in2021 IEEE International Joint Conference on Biometrics (IJCB). IEEE, 2021, pp. 1–8

    2021

  39. [39]

    Mixconv: Mixed depthwise convolutional kernels,

    M. Tan and Q. V . Le, “Mixconv: Mixed depthwise convolutional kernels,” arXiv preprint arXiv:1907.09595, 2019

    work page arXiv 1907

  40. [40]

    Shift: A zero flop, zero parameter alterna- tive to spatial convolutions,

    B. Wu, A. Wan, X. Yue, P. Jin, S. Zhao, N. Golmant, A. Gholaminejad, J. Gonzalez, and K. Keutzer, “Shift: A zero flop, zero parameter alterna- tive to spatial convolutions,” inProceedings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 9127–9135

    2018

  41. [41]

    Shufflefacenet: A lightweight face architecture for efficient and highly-accurate face recog- nition,

    Y . Martindez-Diaz, L. S. Luevano, H. Mendez-Vazquez, M. Nicolas- Diaz, L. Chang, and M. Gonzalez-Mendoza, “Shufflefacenet: A lightweight face architecture for efficient and highly-accurate face recog- nition,” inProceedings of the IEEE/CVF International Conference on Computer Vision Workshops, 2019, pp. 0–0

    2019

  42. [42]

    Shufflenet v2: Practical guidelines for efficient cnn architecture design,

    N. Ma, X. Zhang, H.-T. Zheng, and J. Sun, “Shufflenet v2: Practical guidelines for efficient cnn architecture design,” inProceedings of the European conference on computer vision (ECCV), 2018, pp. 116–131

    2018

  43. [43]

    Pocketnet: Extreme lightweight face recognition network using neural architecture search and multistep knowledge distillation,

    F. Boutros, P. Siebke, M. Klemt, N. Damer, F. Kirchbuchner, and A. Kuijper, “Pocketnet: Extreme lightweight face recognition network using neural architecture search and multistep knowledge distillation,” IEEE Access, vol. 10, pp. 46 823–46 833, 2022

    2022

  44. [44]

    Learning Face Representation from Scratch

    D. Yi, Z. Lei, S. Liao, and S. Z. Li, “Learning face representation from scratch,”arXiv preprint arXiv:1411.7923, 2014

    work page Pith review arXiv 2014

  45. [45]

    Vargfacenet: An efficient variable group convolutional neural network for lightweight face recognition,

    M. Yan, M. Zhao, Z. Xu, Q. Zhang, G. Wang, and Z. Su, “Vargfacenet: An efficient variable group convolutional neural network for lightweight face recognition,” inProceedings of the IEEE/CVF International Confer- ence on Computer Vision Workshops, 2019, pp. 0–0

    2019

  46. [46]

    Lightweight face recognition challenge,

    J. Deng, J. Guo, D. Zhang, Y . Deng, X. Lu, and S. Shi, “Lightweight face recognition challenge,” inProceedings of the IEEE/CVF International Conference on Computer Vision Workshops, 2019, pp. 0–0

    2019

  47. [47]

    Knowledge distillation for face recognition using synthetic data with dynamic latent sampling,

    H. O. Shahreza, A. George, and S. Marcel, “Knowledge distillation for face recognition using synthetic data with dynamic latent sampling,” IEEE Access, 2024

    2024

  48. [48]

    Digi2real: Bridging the realism gap in synthetic data face recognition via foundation models,

    A. George and S. Marcel, “Digi2real: Bridging the realism gap in synthetic data face recognition via foundation models,” inProceedings of the Winter Conference on Applications of Computer Vision, 2025, pp. 1469–1478

    2025

  49. [49]

    Model ru- bik’s cube: Twisting resolution, depth and width for tinynets,

    K. Han, Y . Wang, Q. Zhang, W. Zhang, C. Xu, and T. Zhang, “Model ru- bik’s cube: Twisting resolution, depth and width for tinynets,”Advances in Neural Information Processing Systems, vol. 33, pp. 19 353–19 364, 2020

    2020

  50. [50]

    Ghostfacenets: Lightweight face recognition model from cheap opera- tions,

    M. Alansari, O. A. Hay, S. Javed, A. Shoufan, Y . Zweiri, and N. Werghi, “Ghostfacenets: Lightweight face recognition model from cheap opera- tions,”IEEE Access, 2023

    2023

  51. [51]

    Edgenext: efficiently amalgamated cnn- transformer architecture for mobile vision applications,

    M. Maaz, A. Shaker, H. Cholakkal, S. Khan, S. W. Zamir, R. M. Anwer, and F. Shahbaz Khan, “Edgenext: efficiently amalgamated cnn- transformer architecture for mobile vision applications,” inComputer Vision–ECCV 2022 Workshops: Tel Aviv, Israel, October 23–27, 2022, Proceedings, Part VII. Springer, 2023, pp. 3–20

    2022

  52. [52]

    EFaR 2023: Efficient face recognition competition,

    J. N. Kolf, F. Boutros, J. Elliesen, M. Theuerkauf, N. Damer, M. Alansari, O. A. Hay, S. Alansari, S. Javed, N. Werghiet al., “EFaR 2023: Efficient face recognition competition,” in2023 International Joint Conference on Biometrics (IJCB). IEEE, 2023

    2023

  53. [53]

    Layer Normalization

    J. L. Ba, J. R. Kiros, and G. E. Hinton, “Layer normalization,”arXiv preprint arXiv:1607.06450, 2016

    work page internal anchor Pith review arXiv 2016

  54. [54]

    Image style transfer using convolutional neural networks,

    L. A. Gatys, A. S. Ecker, and M. Bethge, “Image style transfer using convolutional neural networks,” inProceedings of the IEEE conference on computer vision and pattern recognition, 2016, pp. 2414–2423

    2016

  55. [55]

    Tuning layernorm in attention: Towards efficient multi-modal LLM finetuning,

    B. Zhao, H. Tu, C. Wei, J. Mei, and C. Xie, “Tuning layernorm in attention: Towards efficient multi-modal LLM finetuning,” inThe Twelfth International Conference on Learning Representations, 2024. [Online]. Available: https://openreview.net/forum?id=YR3ETaElNK

    2024

  56. [56]

    Lora: Low-rank adaptation of large language models

    E. J. Hu, Y . Shen, P. Wallis, Z. Allen-Zhu, Y . Li, S. Wang, L. Wang, W. Chenet al., “Lora: Low-rank adaptation of large language models.” ICLR, vol. 1, no. 2, p. 3, 2022

    2022

  57. [57]

    Understanding and im- proving layer normalization,

    J. Xu, X. Sun, Z. Zhang, G. Zhao, and J. Lin, “Understanding and im- proving layer normalization,”Advances in neural information processing systems, vol. 32, 2019

    2019

  58. [58]

    Webface260m: A benchmark unveiling the power of million-scale deep face recognition,

    Z. Zhu, G. Huang, J. Deng, Y . Ye, J. Huang, X. Chen, J. Zhu, T. Yang, J. Lu, D. Duet al., “Webface260m: A benchmark unveiling the power of million-scale deep face recognition,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021, pp. 10 492–10 502

    2021

  59. [59]

    Bob: a free signal processing and machine learning toolbox for researchers,

    A. Anjos, L. E. Shafey, R. Wallace, M. G ¨unther, C. McCool, and S. Marcel, “Bob: a free signal processing and machine learning toolbox for researchers,” in20th ACM Conference on Multimedia Systems (ACMMM), Nara, Japan, Oct. 2012. [Online]. Available: https://publications.idiap.ch/downloads/papers/2012/Anjos Bob ACMMM12.pdf

    2012

  60. [60]

    Continuously reproducing toolchains in pattern recognition and machine learning experiments,

    A. Anjos, M. G ¨unther, T. de Freitas Pereira, P. Korshunov, A. Mohammadi, and S. Marcel, “Continuously reproducing toolchains in pattern recognition and machine learning experiments,” in International Conference on Machine Learning (ICML), Aug. 2017. [Online]. Available: http://publications.idiap.ch/downloads/papers/2017/ Anjos ICML2017-2 2017.pdf

    2017

  61. [61]

    A comprehensive database for benchmarking imaging systems,

    K. Panetta, Q. Wan, S. Agaian, S. Rajeev, S. Kamath, R. Rajendran, S. P. Rao, A. Kaszowska, H. A. Taylor, A. Samaniet al., “A comprehensive database for benchmarking imaging systems,”IEEE transactions on pattern analysis and machine intelligence, vol. 42, no. 3, pp. 509–520, 2018

    2018

  62. [62]

    The high-quality wide multi-channel attack (hq-wmca) database,

    Z. Mostaani, A. George, G. Heusch, D. Geissbuhler, and S. Marcel, “The high-quality wide multi-channel attack (hq-wmca) database,”arXiv preprint arXiv:2009.09703, 2020

    work page arXiv 2009

  63. [63]

    A polarimetric thermal database for face recognition research,

    S. Hu, N. J. Short, B. S. Riggan, C. Gordon, K. P. Gurton, M. Thielke, P. Gurram, and A. L. Chan, “A polarimetric thermal database for face recognition research,” inProceedings of the IEEE conference on computer vision and pattern recognition workshops, 2016, pp. 119–126

    2016

  64. [64]

    SCface–surveillance cameras face database,

    M. Grgic, K. Delac, and S. Grgic, “SCface–surveillance cameras face database,”Multimedia tools and applications, vol. 51, no. 3, pp. 863– 879, 2011

    2011

  65. [65]

    Coupled information-theoretic encoding for face photo-sketch recognition,

    W. Zhang, X. Wang, and X. Tang, “Coupled information-theoretic encoding for face photo-sketch recognition,” inCVPR 2011. IEEE, 2011, pp. 513–520

    2011

  66. [66]

    The FERET database and evaluation procedure for face-recognition algorithms,

    P. J. Phillips, H. Wechsler, J. Huang, and P. J. Rauss, “The FERET database and evaluation procedure for face-recognition algorithms,”Im- age and vision computing, vol. 16, no. 5, pp. 295–306, 1998

    1998

  67. [67]

    Identity-aware CycleGAN for face photo-sketch synthesis and recognition,

    Y . Fang, W. Deng, J. Du, and J. Hu, “Identity-aware CycleGAN for face photo-sketch synthesis and recognition,”Pattern Recognition, vol. 102, p. 107249, 2020

    2020

  68. [68]

    The CASIA Nir-Vis 2.0 face database,

    S. Li, D. Yi, Z. Lei, and S. Liao, “The CASIA Nir-Vis 2.0 face database,” inProceedings of the IEEE conference on computer vision and pattern recognition workshops, 2013, pp. 348–353

    2013

  69. [69]

    xEdgeFace: Efficient cross-spectral face recognition for edge devices,

    A. George and S. Marcel, “xEdgeFace: Efficient cross-spectral face recognition for edge devices,” pp. 1–10, 2025

    2025

  70. [70]

    Labeled faces in the wild: A database forstudying face recognition in unconstrained environments,

    G. B. Huang, M. Mattar, T. Berg, and E. Learned-Miller, “Labeled faces in the wild: A database forstudying face recognition in unconstrained environments,” inWorkshop on faces in’Real-Life’Images: detection, alignment, and recognition, 2008

    2008

  71. [71]

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

    T. Zheng, W. Deng, and J. Hu, “Cross-age lfw: A database for study- ing cross-age face recognition in unconstrained environments,”arXiv preprint arXiv:1708.08197, 2017

    work page Pith review arXiv 2017

  72. [72]

    Cross-pose lfw: A database for studying cross- pose face recognition in unconstrained environments,

    T. Zheng and W. Deng, “Cross-pose lfw: A database for studying cross- pose face recognition in unconstrained environments,”Beijing University of Posts and Telecommunications, Tech. Rep, vol. 5, no. 7, 2018

    2018

  73. [73]

    Frontal to profile face verification in the wild,

    S. Sengupta, J.-C. Chen, C. Castillo, V . M. Patel, R. Chellappa, and D. W. Jacobs, “Frontal to profile face verification in the wild,” in2016 IEEE winter conference on applications of computer vision (WACV). IEEE, 2016, pp. 1–9

    2016

  74. [74]

    Agedb: the first manually collected, in-the-wild age database,

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

    2017

  75. [75]

    A light CNN for deep face representation with noisy labels,

    X. Wu, R. He, Z. Sun, and T. Tan, “A light CNN for deep face representation with noisy labels,”IEEE Transactions on Information Forensics and Security, vol. 13, no. 11, pp. 2884–2896, 2018. JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015 12

    2018

  76. [76]

    Dual variational generation for low shot heterogeneous face recognition,

    C. Fu, X. Wu, Y . Hu, H. Huang, and R. He, “Dual variational generation for low shot heterogeneous face recognition,” inAdvances in Neural Information Processing Systems, 2019

    2019

  77. [77]

    Heterogeneous face recognition using inter-session variability modelling,

    T. de Freitas Pereira and S. Marcel, “Heterogeneous face recognition using inter-session variability modelling,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops, 2016, pp. 111–118

    2016

  78. [78]

    Cross-eyed 2017: Cross-spectral iris/periocular recognition com- petition,

    A. F. Sequeira, L. Chen, J. Ferryman, P. Wild, F. Alonso-Fernandez, J. Bigun, K. B. Raja, R. Raghavendra, C. Busch, T. de Freitas Pereira et al., “Cross-eyed 2017: Cross-spectral iris/periocular recognition com- petition,” in2017 IEEE International Joint Conference on Biometrics (IJCB). IEEE, 2017, pp. 725–732

    2017

  79. [79]

    The facesketchid system: Matching facial composites to mugshots,

    S. J. Klum, H. Han, B. F. Klare, and A. K. Jain, “The facesketchid system: Matching facial composites to mugshots,”IEEE Transactions on Information Forensics and Security, vol. 9, no. 12, pp. 2248–2263, 2014

    2014

  80. [80]

    Seeing the forest from the trees: A holistic approach to near-infrared heterogeneous face recognition,

    C. Reale, N. M. Nasrabadi, H. Kwon, and R. Chellappa, “Seeing the forest from the trees: A holistic approach to near-infrared heterogeneous face recognition,” inIEEE Conference on Computer Vision and Pattern Recognition Workshops, 2016

    2016

Showing first 80 references.