pith. machine review for the scientific record. sign in

arxiv: 2604.17321 · v1 · submitted 2026-04-19 · 💻 cs.CV

Recognition: unknown

R-FLoRA: Residual-Statistic-Gated Low-Rank Adaptation for Single-Image Face Morphing Attack Detection

Raghavendra Ramachandra
Authors on Pith no claims yet

Pith reviewed 2026-05-10 06:53 UTC · model grok-4.3

classification 💻 cs.CV
keywords face morphing attack detectionsingle-image MADlow-rank adaptationresidual statisticsvision transformerLaplacian residualsbiometric securitycross-domain generalization
0
0 comments X

The pith

Residual-statistic-gated low-rank adapters let a frozen vision transformer detect single-image face morphing attacks more accurately than prior methods.

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

The paper proposes R-FLoRA, a framework that combines high-frequency Laplacian residual statistics with a frozen vision transformer to spot morphing artifacts in single face images. Low-rank adapters are gated by these residuals, fused via Res-FiLM, and regularized with a residual-contrastive alignment loss to sharpen local artifact detection without retraining the backbone. Tests on four ICAO-compliant datasets with seven morph generation techniques show the method beats nine recent S-MAD algorithms in accuracy and cross-domain generalization. This matters for biometric systems because single-image detection is required when no trusted reference is available, such as in passport issuance. The design keeps the model efficient and interpretable for practical deployment.

Core claim

The R-FLoRA framework integrates high-frequency Laplacian residual statistics with representations from a frozen foundation-scale vision transformer through residual-statistic-gated low-rank adapters and feature-wise residual fusion (Res-FiLM), further regularized by a residual-contrastive alignment loss. This enhances sensitivity to local morphing artifacts in single facial images while preserving semantic context, yielding consistent improvements in detection accuracy and cross-domain generalization over nine state-of-the-art S-MAD algorithms on four ICAO-compliant datasets encompassing seven morph generation techniques.

What carries the argument

R-FLoRA (residual-statistic-gated low-rank adapters) that use high-frequency Laplacian residuals to control adapter behavior, together with Res-FiLM fusion and the residual-contrastive alignment loss, to direct the frozen transformer toward morphing artifacts.

Load-bearing premise

High-frequency Laplacian residual statistics reliably capture morphing artifacts across diverse unseen generation methods, and the residual-contrastive alignment loss improves discrimination without introducing dataset-specific biases or overfitting.

What would settle it

A new test set using an eighth morph generation technique not among the seven evaluated, where the method's accuracy falls below the best prior S-MAD algorithm, would falsify the generalization claim.

Figures

Figures reproduced from arXiv: 2604.17321 by Raghavendra Ramachandra.

Figure 1
Figure 1. Figure 1: Overall architecture of the proposed residual-conditioned morphing attack detection framework. The input image is [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Example facial images from the four different face morphing datasets with seven different Morphing Types (MT) (a) [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Examples of Gaussian blur degradation applied at test [PITH_FULL_IMAGE:figures/full_fig_p012_3.png] view at source ↗
Figure 5
Figure 5. Figure 5: Examples of JPEG compression applied at test time [PITH_FULL_IMAGE:figures/full_fig_p013_5.png] view at source ↗
read the original abstract

Face morphing attacks pose a substantial risk to the reliability of face recognition systems used in passport issuance, border control, and digital identity verification. Detecting morphing attacks from a single facial image remains challenging owing to the lack of a trusted reference and the diversity of attack generation methods. This paper presents a new Single-Image Face Morphing Attack Detection (S-MAD) framework that integrates high-frequency Laplacian residual statistics with representations from a frozen, foundation-scale vision transformer. The approach employs residual-statistic-gated low-rank adapters (R-FLoRA) and feature-wise residual fusion (Res-FiLM) to enhance sensitivity to local morphing artefacts while preserving the semantic context of the backbone. A novel residual-contrastive alignment loss further regularises the fused token space, improving discrimination under unseen morphing conditions. Comprehensive experiments on four ICAO-compliant datasets, encompassing seven morph generation techniques, demonstrate that the proposed method consistently surpasses nine recent state-of-the-art S-MAD algorithms in detection accuracy and cross-domain (or dataset) generalisation. With a frozen backbone and minimal trainable parameters, the model achieves real-time efficiency and interpretability, making it suitable for real-life scenarios in biometric verification systems.

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 / 2 minor

Summary. The paper proposes R-FLoRA, a single-image face morphing attack detection (S-MAD) framework that fuses high-frequency Laplacian residual statistics with a frozen foundation-scale vision transformer via residual-statistic-gated low-rank adapters (R-FLoRA) and Res-FiLM feature-wise fusion. A residual-contrastive alignment loss is introduced to regularize the token space. Experiments across four ICAO-compliant datasets spanning seven morph generation techniques claim consistent outperformance of nine recent SOTA S-MAD methods in accuracy and cross-domain generalization, while using minimal trainable parameters for real-time efficiency and interpretability.

Significance. If the performance and generalization results hold under rigorous validation, the work would meaningfully advance practical biometric security by offering an efficient, low-parameter S-MAD solution suitable for real-world passport and border systems. The frozen backbone plus gated adapters and explicit residual handling provide a principled way to inject artefact sensitivity without full fine-tuning, addressing the core difficulty of unseen morphing pipelines. Strengths include the emphasis on cross-dataset evaluation and the interpretability angle.

major comments (2)
  1. [Abstract and §4 (Experiments)] Abstract and §4 (Experiments): The central claim of consistent superiority and cross-domain generalization over nine SOTA algorithms is presented without statistical significance tests, error bars or standard deviations across runs, or explicit descriptions of the train/test splits and cross-dataset protocols. This absence prevents verification that the reported gains are robust rather than artifacts of particular partitions or evaluation choices.
  2. [§3 (Method)] §3 (Method): The load-bearing assumption that high-frequency Laplacian residuals plus the residual-contrastive alignment loss isolate universal morphing artefacts (rather than dataset-specific high-frequency noise or acquisition conditions) lacks supporting analysis. No frequency spectra, artefact localization maps, or ablation isolating the residuals' contribution across the seven morph techniques are provided to confirm that the gated adapters focus on blending boundaries consistently in unseen conditions.
minor comments (2)
  1. [Abstract] The abstract introduces R-FLoRA without spelling out the acronym on first use, although the title makes the expansion clear.
  2. [§2 (Related Work)] Related-work section could more explicitly contrast the proposed residual gating against prior frequency-based or adapter-based MAD methods to sharpen the novelty claim.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback. We address each major comment point by point below, indicating the revisions we will make to improve the manuscript.

read point-by-point responses

  1. Referee: [Abstract and §4 (Experiments)] Abstract and §4 (Experiments): The central claim of consistent superiority and cross-domain generalization over nine SOTA algorithms is presented without statistical significance tests, error bars or standard deviations across runs, or explicit descriptions of the train/test splits and cross-dataset protocols. This absence prevents verification that the reported gains are robust rather than artifacts of particular partitions or evaluation choices.

    Authors: We agree that the absence of statistical significance tests, error bars, and explicit protocol details limits the ability to fully verify robustness. In the revised manuscript we will add standard deviations and error bars computed over multiple random seeds, p-values for key comparisons against the nine baselines, and expanded descriptions of the train/test splits together with the precise cross-dataset protocols used in §4. These additions will directly substantiate the claims of consistent superiority and generalization. revision: yes

  2. Referee: [§3 (Method)] §3 (Method): The load-bearing assumption that high-frequency Laplacian residuals plus the residual-contrastive alignment loss isolate universal morphing artefacts (rather than dataset-specific high-frequency noise or acquisition conditions) lacks supporting analysis. No frequency spectra, artefact localization maps, or ablation isolating the residuals' contribution across the seven morph techniques are provided to confirm that the gated adapters focus on blending boundaries consistently in unseen conditions.

    Authors: The reported cross-dataset results across four ICAO-compliant datasets and seven distinct morph generation techniques already supply empirical evidence that the captured artefacts generalize beyond dataset-specific noise. To address the request for direct supporting analysis, the revised version will incorporate frequency spectra comparisons, artefact localization maps extracted from the gated adapters, and targeted ablations that isolate the Laplacian residual statistics and residual-contrastive loss across each of the seven morph techniques. These additions will more explicitly demonstrate consistent focus on blending boundaries under unseen conditions. revision: yes

Circularity Check

0 steps flagged

Empirical framework with no detectable circularity

full rationale

The paper describes an empirical S-MAD method that combines high-frequency Laplacian residuals with a frozen ViT backbone via R-FLoRA adapters and Res-FiLM fusion, plus a residual-contrastive loss, and reports accuracy on four external ICAO datasets covering seven morph techniques. No equations, derivations, or self-referential definitions appear in the provided text; performance claims rest on cross-dataset experimental comparisons rather than any reduction of outputs to fitted inputs or self-citations by construction. The derivation chain is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Review is abstract-only; no mathematical derivations, free parameters, axioms, or new entities are described. The approach builds on standard components (frozen ViT, LoRA adapters, contrastive loss) whose assumptions are not detailed here.

pith-pipeline@v0.9.0 · 5512 in / 1360 out tokens · 56033 ms · 2026-05-10T06:53:01.083793+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

43 extracted references · 7 canonical work pages · 2 internal anchors

  1. [1]

    Face morphing attack generation and detection: A comprehensive survey,

    S. Venkatesh, R. Raghavendra, K. Raja, and C. Busch, “Face morphing attack generation and detection: A comprehensive survey,”IEEE Trans- actions on Technology and Society, vol. 2, no. 3, pp. 128–145, March 2021

    2021

  2. [2]

    Humans vs. algorithms: Assessment of security risks posed by facial morphing to identity verification at border control,

    A. Makrushin., T. Neubert., and J. Dittmann., “Humans vs. algorithms: Assessment of security risks posed by facial morphing to identity verification at border control,” inProceedings of the 14th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications - Volume 4: VISAPP ,, INSTICC. SciTePress, 2019, pp. 513–520

    2019

  3. [3]

    Simulation of border control in an ongoing web-based experiment for estimating morphing detection performance of humans,

    A. Makrushin, D. Siegel, and J. Dittmann, “Simulation of border control in an ongoing web-based experiment for estimating morphing detection performance of humans,” inProceedings of the 2020 ACM Workshop on Information Hiding and Multimedia Security, ser. IH&MMSec ’20. New York, NY , USA: Association for Computing Machinery, 2020, p. 91–96. [Online]. Avai...

    work page doi:10.1145/3369412.3395073 2020

  4. [4]

    Analyzing human observer ability in morphing attack detection—where do we stand?

    S. R. Godage, F. Løv ˚asdal, S. Venkatesh, K. Raja, R. Ramachandra, and C. Busch, “Analyzing human observer ability in morphing attack detection—where do we stand?”IEEE Transactions on Technology and Society, vol. 4, no. 2, pp. 125–145, 2023

    2023

  5. [5]

    Detecting morphed face images,

    R. Raghavendra, K. Raja, and C. Busch, “Detecting morphed face images,” in2016 IEEE 8th Intl. Conf. on Biometrics: Theory, Applica- tions and Systems (BTAS), 8th IEEE Intl. Conf. on Biometrics: Theory, Applications and Systems (BTAS-2016). IEEE, September 2016

    2016

  6. [6]

    Transferable deep-CNN features for detecting digital and print-scanned morphed face images,

    R. Raghavendra, K. Raja, S. Venkatesh, and C. Busch, “Transferable deep-CNN features for detecting digital and print-scanned morphed face images,” inIEEE Conf. on Computer Vision and Pattern Recognition Workshops (CVPRW), 2017, pp. 1822–1830

    2017

  7. [7]

    On the generalization of detecting face morphing attacks as anomalies: Novelty vs. outlier detection,

    N. Damer, J. H. Grebe, S. Zienert, F. Kirchbuchner, and A. Kuijper, “On the generalization of detecting face morphing attacks as anomalies: Novelty vs. outlier detection,” in2019 IEEE 10th International Confer- ence on Biometrics Theory, Applications and Systems (BTAS), 2019, pp. 1–5

    2019

  8. [8]

    PW-MAD: Pixel-wise supervision for generalized face morphing attack detection,

    N. Damer, N. Spiller, M. Fang, F. Boutros, F. Kirchbuchner, and A. Kuijper, “PW-MAD: Pixel-wise supervision for generalized face morphing attack detection,” inAdvances in Visual Computing, G. Bebis, V . Athitsos, T. Yan, M. Lau, F. Li, C. Shi, X. Yuan, C. Mousas, and G. Bruder, Eds. Cham: Springer International Publishing, 2021, pp. 291–304. ACCEPTED IN I...

    2021

  9. [9]

    Multimodality for reliable single image based face morphing attack detection,

    R. Ramachandra and G. Li, “Multimodality for reliable single image based face morphing attack detection,”IEEE Access, vol. 10, pp. 82 418– 82 433, 2022

    2022

  10. [10]

    Unsupervised face morphing attack detection via self-paced anomaly detection,

    M. Fang, F. Boutros, and N. Damer, “Unsupervised face morphing attack detection via self-paced anomaly detection,” in2022 IEEE International Joint Conference on Biometrics (IJCB). IEEE, 2022, pp. 1–11

    2022

  11. [11]

    Detecting morphed face attacks using residual noise from deep multi-scale context aggregation network,

    S. Venkatesh, R. Raghavendra, K. Raja, L. Spreeuwers, R. Veldhuis, and C. Busch, “Detecting morphed face attacks using residual noise from deep multi-scale context aggregation network,” in2020 IEEE Winter Conference on Applications of Computer Vision (WACV). IEEE, March 2020, pp. 269–278

    2020

  12. [12]

    Effectiveness of residual noise based methods for single image based morphing attack detection: A comparative study,

    S. Patwardhan, S. Venkatesh, and R. Ramachandra, “Effectiveness of residual noise based methods for single image based morphing attack detection: A comparative study,” inAdvances in Visual Computing. Cham: Springer Nature Switzerland, 2025, pp. 315–329

    2025

  13. [13]

    Biometric systems under morphing attacks: Assessment of morphing techniques and vulnerability reporting,

    U. Scherhag and et al., “Biometric systems under morphing attacks: Assessment of morphing techniques and vulnerability reporting,” inIntl. Conf. of the Biometrics Special Interest Group BIOSIG 2017, 2017, pp. 1–7

    2017

  14. [14]

    Empowering mor- phing attack detection using interpretable image-text foundation model,

    S. Patwardhan, R. Ramachandra, and S. Venkatesh, “Empowering mor- phing attack detection using interpretable image-text foundation model,” inComputer Vision and Image Processing. Singapore: Springer Singapore, 2024, pp. 163–169

    2024

  15. [15]

    Foundpad: Foundation models reloaded for face presentation attack detection,

    G. Ozgur, E. Caldeira, T. Chettaoui, F. Boutros, R. Ramachandra, and N. Damer, “Foundpad: Foundation models reloaded for face presentation attack detection,” inWACV AI4MFDD Workshop, 2025. [Online]. Available: https://arxiv.org/abs/2501.02892

    work page arXiv 2025

  16. [16]

    Color residual noise patterns as deep features for single-image morphing attack detection,

    N. S. S. R. R. S. V . G. Jaswal, “Color residual noise patterns as deep features for single-image morphing attack detection,” inComputer Vision and Image Processing. Singapore: Springer Singapore, 2025, pp. 180– 192

    2025

  17. [17]

    Deep learning techniques for detecting morphed face images: A literature review,

    P. Morais, I. Domingues, and J. Bernardino, “Deep learning techniques for detecting morphed face images: A literature review,”IEEE Access, vol. 13, pp. 105 952–105 981, 2025

    2025

  18. [18]

    Robust face morphing attack detection using fusion of multiple features and classification techniques,

    J. M. Singh, S. Venkatesh, and R. Ramachandra, “Robust face morphing attack detection using fusion of multiple features and classification techniques,” in2023 26th International Conference on Information Fusion (FUSION), 2023, pp. 1–8

    2023

  19. [19]

    Face morphing attack detection with denoising diffusion probabilistic models,

    M. Ivanovska and V . ˇStruc, “Face morphing attack detection with denoising diffusion probabilistic models,” in2023 11th International Workshop on Biometrics and Forensics (IWBF), 2023

    2023

  20. [20]

    Alphanet: Single morphing attack detection using multiple contributors,

    J. Tapia and C. Busch, “Alphanet: Single morphing attack detection using multiple contributors,” in2023 IEEE International Workshop on Information Forensics and Security (WIFS). IEEE, 2023, pp. 1–6

    2023

  21. [21]

    Single-morphing attack detection using few-shot learning and triplet-loss,

    J. E. Tapia, D. Schulz, and C. Busch, “Single-morphing attack detection using few-shot learning and triplet-loss,”Neurocomputing, vol. 636, p. 130033, 2025

    2025

  22. [22]

    Madation: Face morphing attack detection with foundation models,

    E. Caldeira, G. Ozgur, T. Chettaoui, M. Ivanovska, P. Peer, F. Boutros, V . Struc, and N. Damer, “Madation: Face morphing attack detection with foundation models,” 2025. [Online]. Available: https://arxiv.org/abs/2501.03800

    work page arXiv 2025

  23. [23]

    Learning transferable visual models from natural language supervision,

    A. Radford, J. W. Kim, C. Hallacy, A. Ramesh, G. Goh, S. Agarwal, G. Sastry, A. Askell, P. Mishkin, J. Clarket al., “Learning transferable visual models from natural language supervision,” inInternational conference on machine learning. PmLR, 2021, pp. 8748–8763

    2021

  24. [24]

    DINOv2: Learning Robust Visual Features without Supervision

    M. Oquab, T. Darcet, T. Moutakanni, H. V o, M. Szafraniec, V . Khalidov, P. Fernandez, D. Haziza, F. Massa, A. El-Noubyet al., “Dinov2: Learning robust visual features without supervision,”arXiv preprint arXiv:2304.07193, 2023

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  25. [25]

    DINOv3

    O. Sim ´eoni, H. V . V o, M. Seitzer, F. Baldassarre, M. Oquab, C. Jose, V . Khalidov, M. Szafraniec, S. Yi, M. Ramamonjisoaet al., “Dinov3,” arXiv preprint arXiv:2508.10104, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  26. [26]

    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,”Image and vision computing, vol. 16, no. 5, pp. 295–306, 1998

    1998

  27. [27]

    Face detection algorithm and feature performance on frgc 2.0 imagery,

    J. R. Beveridge, A. Alvarez, J. Saraf, W. Fisher, P. J. Flynn, and J. Gentile, “Face detection algorithm and feature performance on frgc 2.0 imagery,” in2007 First IEEE International Conference on Biometrics: Theory, Applications, and Systems, 2007, pp. 1–7

    2007

  28. [28]

    Audioclip: Extending clip to image, text and audio

    E. Sarkar, P. Korshunov, L. Colbois, and S. Marcel, “Are gan-based morphs threatening face recognition?” inICASSP 2022 - 2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2022, pp. 2959–2963. [Online]. Available: https://doi.org/10.1109/ICASSP43922.2022.9746477

    work page doi:10.1109/icassp43922.2022.9746477 2022

  29. [29]

    Mutual information based multispectral image fusion for improved face recognition,

    R. Raghavendra, S. Venkatesh, K. B. Raja, F. A. Cheikh, and C. Busch, “Mutual information based multispectral image fusion for improved face recognition,” in2016 12th International Conference on Signal-Image Technology & Internet-Based Systems (SITIS), 2016, pp. 62–68

    2016

  30. [30]

    Machine readable passports – part 1 – introduction,

    I. C. A. Organization, “Machine readable passports – part 1 – introduction,” http://www.icao.int/publications/Documents/9303 p1 cons en.pdf, International Civil Aviation Organization (ICAO), 2015

    2015

  31. [31]

    Face morphing versus face averaging: Vulnerability and detection,

    R. Raghavendra, K. Raja, S. Venkatesh, and C. Busch, “Face morphing versus face averaging: Vulnerability and detection,” inProc. Intl. Joint Conf. on Biometrics (IJCB), 2017

    2017

  32. [32]

    MagFace: A universal representation for face recognition and quality assessment,

    Q. Meng, S. Zhao, Z. Huang, and F. Zhou, “MagFace: A universal representation for face recognition and quality assessment,” 2021

    2021

  33. [33]

    Magic morph 1.95,

    “Magic morph 1.95,” https://downloads.tomsguide.com/magic-morph, 0301-6817.html, 2020, accessed: October 2020

    2020

  34. [34]

    MIPGAN—Generating strong and high quality morphing attacks using identity prior driven GAN,

    H. Zhang, S. Venkatesh, R. Raghavendra, K. Raja, N. Damer, and C. Busch, “MIPGAN—Generating strong and high quality morphing attacks using identity prior driven GAN,”IEEE Transactions on Biomet- rics, Behavior, and Identity Science, vol. 3, no. 3, pp. 365–383, 2021

    2021

  35. [35]

    Mordiff: Recognition vulnerability and attack detectability of face morphing attacks created by diffusion autoencoders,

    N. Damer, M. Fang, P. Siebke, J. N. Kolf, M. Huber, and F. Boutros, “Mordiff: Recognition vulnerability and attack detectability of face morphing attacks created by diffusion autoencoders,” in2023 11th International Workshop on Biometrics and Forensics (IWBF), 2023, pp. 1–6

    2023

  36. [36]

    Morph-pipe: Plugging in identity prior to enhance face morphing attack based on diffusion model,

    H. Zhang, R. Ramachandra, K. Raja, and C. Busch, “Morph-pipe: Plugging in identity prior to enhance face morphing attack based on diffusion model,” inNorsk IKT-konferanse for forskning og utdanning (NISK), vol. 3, 2023, pp. 1–6

    2023

  37. [37]

    Greedy-dim: Greedy algorithms for unreasonably effective face morphs,

    Z. W. Blasingame and C. Liu, “Greedy-dim: Greedy algorithms for unreasonably effective face morphs,” in2024 IEEE International Joint Conference on Biometrics (IJCB), 2024, pp. 1–11

    2024

  38. [38]

    Information Tech- nology - Biometric presentation attack detection - Part 3: Testing and Reporting, International Organization for Standardization, 2017

    ISO/IEC JTC1 SC37 Biometrics,ISO/IEC 30107-3. Information Tech- nology - Biometric presentation attack detection - Part 3: Testing and Reporting, International Organization for Standardization, 2017

    2017

  39. [39]

    Selfmad: Enhancing generalization and robustness in mor- phing attack detection via self-supervised learning,

    M. Ivanovska, L. Todorov, N. Damer, D. K. Jain, P. Peer, and V . ˇStruc, “Selfmad: Enhancing generalization and robustness in mor- phing attack detection via self-supervised learning,”arXiv preprint arXiv:2504.05504, 2025

    work page arXiv 2025

  40. [40]

    Face morphing attack detection and localization based on feature-wise supervision,

    L. Qin, F. Peng, and M. Long, “Face morphing attack detection and localization based on feature-wise supervision,”IEEE Transactions on Information Forensics and Security, vol. 17, pp. 3649–3662, 2022

    2022

  41. [41]

    Face morphing detection using fourier spectrum of sensor pattern noise,

    L.-B. Zhang, F. Peng, and M. Long, “Face morphing detection using fourier spectrum of sensor pattern noise,” in2018 IEEE international conference on multimedia and expo (ICME). IEEE, 2018, pp. 1–6

    2018

  42. [42]

    Lora-enhanced vision transformer for single image based morphing at- tack detection via knowledge distillation from efficientnet,

    S. Ria, P. Sushrut, R. Raghavendra, C. Praveen Kumar, and U. Kishor, “Lora-enhanced vision transformer for single image based morphing at- tack detection via knowledge distillation from efficientnet,” inComputer Vision and Image Processing. Singapore: Springer Singapore, 2025, pp. 163–169

    2025

  43. [43]

    Generalized single- image-based morphing attack detection using deep representations from vision transformer,

    H. Zhang, R. Ramachandra, K. Raja, and C. Busch, “Generalized single- image-based morphing attack detection using deep representations from vision transformer,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 1510–1518

    2024