Pith. sign in

REVIEW 3 major objections 6 minor 89 references

Fairer Analysis and Demographically Balanced Face Generation for Fairer Face Verification

T0 review · 3 major / 6 minor · reviewed 2026-08-11 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Balancing demographic attributes inside a synthetic face-generation pipeline reduces verification bias more than resampling or loss weighting, with a slight accuracy gain.

desk verdict Useful controlled generation recipe and a genuinely useful logit/ANOVA fairness analysis, but the paper conflates balancing with a style-matching fix, so the central attribution needs a matched-but-imbalanced control. read the letter →

arxiv 2412.03349 v2 pith:DQHQ5KZ6 submitted 2024-12-04 cs.CV

Alexandre Fournier-Montgieux , Michael Soumm , Adrian Popescu , Bertrand Luvison , Hervé Le Borgne This is my paper · ORCID
classification cs.CV
keywords faceverificationfairnesssyntheticdatadiffusionmodelsdemographicbiasDCFacelogitregressionANOVA
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper claims that face-verification models become fairer when the synthetic training data they are trained on is generated under explicit demographic control, rather than merely re-balanced by resampling or loss weighting. Building on the DCFace dual-condition diffusion pipeline, the authors create two datasets, DCFace+Cge and DCFace+Call, in which gender, ethnicity, age, and pose are balanced during generation. On the RFW, FAVCI2D, and BFW verification benchmarks, models trained on these sets show lower demographic-parity and equalized-odds gaps than models trained on the original DCFace set or on CASIA with standard debiasing, while micro-average accuracy stays roughly level or slightly improves. The paper also introduces a statistical analysis based on logit regression and ANOVA to separate the contribution of each attribute, arguing that this analysis reveals biases that classical fairness metrics miss.

What carries the argument

The load-bearing object is the controlled generation pipeline added on top of DCFace's dual-condition diffusion model. DCFace produces each synthetic identity by applying the style of a real image to a DDPM-generated identity image; the paper gains control by selecting the identity images so that the joint gender×ethnicity distribution is exactly balanced, then diversifying age and pose by repeatedly filling the least-represented categories and requiring ID and style images to belong to the same demographic segment. The second piece of machinery is the statistical analysis: logit regression on true-match and false-match outcomes (with dummy-coded ethnicity and gender, continuous age and pose) gives marginal effects of each attribute on FMR and TMR holding the others constant, and ANOVA on the latent-space distances of positive and negative pairs partitions variance into $\eta^2$ contributions per attribute. Together these tools convert a dataset-level fairness comparison into per-attribute effect sizes.

What would settle it

Compute the fairness metrics and logit marginal effects on an evaluation set with ground-truth demographic labels (e.g., BFW's supplied ethnicity or a manually re-annotated subset of RFW) and compare them to the same metrics computed with FairFace-inferred labels; if the DCFace+Call advantage shrinks or reverses under true labels, the central claim fails.

Watch

Extended reading notes

Core claim

On its own terms, the paper's central discovery is that the distribution of sensitive attributes inside a synthetic training set is a controllable and effective lever for fairness in face verification. The proposed pipeline selects DDPM-generated identity images whose gender×ethnicity distribution is perfectly balanced, and then iteratively populates underrepresented age and pose categories while matching the demographic segment of ID and style images to help the diffusion model converge. Trained with the AdaFace loss on a ResNet50, models built on DCFace+Call raise the equalized-odds ratio on RFW from 15.3 for DCFace to 45.9, and cut DPD from 17.2 to 11.2, while micro-average accuracy rises from 75.6 to 77.3. The logit marginal effects show the African-subgroup FMR penalty drops from 35 points with DCFace to 12 points with the controlled set, while resampling only reduced it to 22. ANOVA on negative-pair latent distances shows ethnicity explains less variance with the balanced sets, indicating the bias reduction is visible in the geometry of the embedding space, not only in the final error rates.

Load-bearing premise

The whole comparison rests on the demographic labels produced by FairFace being accurate enough; for some groups FairFace is only about 58% accurate, so label mistakes could make the balanced pipeline look fairer than it really is.

Editorial extensions

If this is right

  • Models trained on DCFace+Call improve fairness metrics on all three evaluation benchmarks while gaining a small amount of micro-average accuracy over the original DCFace set.
  • The controlled generation reduces the ethnicity-related false-match penalty more than resampling or loss weighting, and avoids the side effect of resampling that increases gender bias while reducing ethnicity bias.
  • ANOVA results show that balancing changes the latent space itself, not just the final threshold decisions, since demographic attributes explain less variance in the distances between negative pairs.
  • The balancing pipeline is presented as adaptable to other synthetic face generators, such as IDiff-Face, so the fairness mechanism may transfer beyond DCFace.
  • The logit and ANOVA approach can serve as a general audit tool for face-verification models, quantifying per-attribute bias even when classical fairness metrics give an incomplete or surprising picture.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Because FairFace labels drive both dataset construction and fairness evaluation, an independent test on manually annotated data would show whether part of the measured gain is an artifact of correlated label errors.
  • The balancing logic is attribute-agnostic and could be applied to other sensitive or nuisance attributes, such as head pose, glasses, or lighting, making the method a template for controlled generation beyond the four attributes tested.
  • The logit/ANOVA toolkit could be used to audit deployed verification systems without retraining, by estimating per-group false-match effects directly from a labeled probe set.
  • The slight accuracy gain on balanced benchmarks suggests that demographic balancing also reduces train-to-evaluation distribution shift, an effect worth measuring on additional real-world protocols.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 6 minor

Summary. The paper proposes two synthetic face training sets, DCFace+Cge and DCFace+Call, built by extending the DCFace dual-conditioned diffusion pipeline with demographic balancing of ID images (gender and ethnicity, and additionally age and pose for Call) and with style images matched to the demographic segment of the ID image. Models trained on these sets are compared, on RFW, FAVCI2D, and BFW, against models trained on real and synthetic baselines, with and without resampling or loss-weighting debiasing. Fairness is measured with DoB, DPD/DPR, EOD/EOR, and with a proposed logit-regression and ANOVA analysis that quantifies attribute effects on verification outcomes and on latent-space distances. The paper reports that the proposed controlled generation improves fairness relative to the other synthetic-data approaches while slightly improving raw accuracy.

Significance. If the attribution were fully supported, this would be a useful contribution: it demonstrates practical control over demographic composition in synthetic face generation and introduces a more granular statistical toolkit for fairness analysis than headline metrics alone. Strengths include the release of code and data, evaluation on three fairness-oriented verification benchmarks against several baselines, and a statistical analysis (logit marginal effects plus ANOVA with diagnostic checks) that goes beyond aggregate scores. The central caveat is that the main comparison conflates demographic balancing with a style-ID matching/curation step, so the specific benefit of balancing is not isolated; in addition, the headline fairness numbers lack uncertainty estimates.

major comments (3)
  1. [Section 3.2 and Supplementary A] The proposed DCFace+Cge and DCFace+Call differ from the DCFace baseline by two simultaneous changes: (i) ID images are selected to balance gender and ethnicity (and age/pose for Call), and (ii) style images are matched to the same gender-by-ethnicity segment as the ID image. Supplementary A states that random style sampling 'results in a non-decreasing loss of the ResNet network' and that matching was introduced because convergence is not guaranteed without it. The baseline DCFace and the DCFace+Sall/DCFace+Wall comparisons use the original unmatched pipeline, so the reported fairness gains cannot be cleanly attributed to demographic balancing; they may come largely from the style-ID matching/curation step. The Cge-versus-Call comparison is not affected by this confound, but the headline DCFace-versus-DCFace+C comparisons are. Please add a matched-but-imbalanced control (DCFace with segment matching but no demographic balancing) or otherwise disentangle the two changes before claiming that balancing is the cause of the fairness improvement.
  2. [Table 2 and Figures 5-6] All fairness metrics and accuracy values are reported for a single training run, with no confidence intervals, bootstrap intervals, or multiple seeds. Fairness metrics such as EOR and EOD are sensitive to pair sampling, and the paper itself notes in Supplementary F that BFW contains very few identities, which can make estimates unstable. Consequently, statements such as 'significantly improves fairness' (Abstract) and 'substantially improves fairness metrics' (Section 5.1) are not yet supported by the evidence. Please provide uncertainty estimates, at least for the key RFW/FAVCI2D/BFW comparisons, and preferably train with multiple seeds to assess variance.
  3. [Sections 3.1 and 4.1, Supplementary Table 4] The training-set balancing and part of the evaluation rely on demographic labels inferred by FairFace, whose per-group accuracy is as low as 0.581 (Latino-Hispanic) and 0.631 (Middle-Eastern) on the FairFace validation set. For FAVCI2D, ethnicity is inferred with FairFace, and age and pose are inferred for all evaluation sets; the same attribute-inference approach is used to construct the balanced training sets. This creates a risk that systematic label errors align generation control with evaluation in a way that inflates the apparent benefit of the proposed method. The limitation is acknowledged in Section 6, but a quantitative sensitivity analysis (e.g., reporting results on subsets with high FairFace confidence, or presenting RFW/BFW ground-truth ethnicity results separately from FairFace-inferred attributes) would allow the reader to assess the magnitude of this effect.
minor comments (6)
  1. [Notation throughout] The naming of the proposed datasets is inconsistent (e.g., 'DCFace + Cge', 'DCFace + C ge', 'DCFace+Cge'); please unify the notation in the text, tables, and figures.
  2. [Table 2] The header for the last fairness column reads 'Equalized Odds Ratio; Acc' but omits the abbreviation EOR; also, the dataset name 'FAVCI2D' is written with inconsistent spacing across the paper.
  3. [Figure 5] Non-significant marginal effects are shown in transparency, but the significance threshold and the method used to compute p-values are not stated; please specify them.
  4. [Section 4.3] In the ANOVA discussion, the sentence 'the total R2 = 0.18 of the ANOVA' does not make clear whether this value refers to positive pairs, negative pairs, or the pooled analysis; please clarify.
  5. [Section 4.2] The definitions of DPD and DPR refer to 'the probability for individuals to receive a positive outcome', but in the verification setting it is not clear whether this probability is TMR, FMR, or a combined accuracy; please define the outcome precisely.
  6. [Supplementary A] The discussion about untested sampling strategies and missing files in the original DCFace code is more appropriate for a reproducibility note than for the main pipeline description; consider moving it to a clearly marked implementation-details section.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the fairness improvements are established on held-out evaluation sets independent of the training-set construction.

full rationale

The paper's central claim is that a controlled generation pipeline (DCFace+Cge/Call) improves fairness of face verification models. The derivation chain is experimental rather than mathematical: the balanced training sets are constructed by selecting DDPM ID images and matching style images using FairFace-inferred attributes, and the resulting models are evaluated on held-out verification datasets (RFW, FAVCI2D, BFW) that are not used to construct the training sets. The fairness metrics (DoB, DPD, EOD, DPR, EOR) are computed on these held-out sets using dataset-provided ethnicity/gender labels where available (RFW and BFW ethnicity; BFW gender), so the evaluation does not reduce to the training-set construction. The logit-regression and ANOVA analyses are post-hoc descriptive tools fitted to evaluation outcomes; they do not feed back into the generation pipeline and are not used to predict the very data from which they were fit. The only measurement-validity concern is that FAVCI2D ethnicity (and age/pose on all sets) are inferred with FairFace, the same tool used to balance the training sets, which could in principle align label errors in both construction and evaluation; however, the central fairness improvements also appear on RFW and BFW, where ethnicity and gender labels are provided with the dataset, so the claim does not reduce by construction. Self-citations ([20,21,53]) are contextual and not load-bearing for the main result. No fitted parameter is renamed as a prediction.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central claim rests on the accuracy of FairFace attribute inference, the stability of the DCFace generation pipeline, and the assumption that the four chosen attributes are the dominant bias factors. The logit model also carries standard statistical assumptions.

free parameters (3)
  • Similarity threshold for filtering DDPM images = 0.6
    The paper sets the cosine similarity threshold to 0.6 when filtering images with the Feval network (supplementary A). This is a hand-picked value that affects which ID images are used, and the paper notes the threshold can vary depending on the Feval training.
  • Demographic segment matching rule = same gender and ethnicity
    The authors match the demographic segment of ID and style images because initial tests showed that mismatching leads to non-convergence of the ResNet50 (Section 3.2, supplementary A). This is an ad hoc design choice that shapes the generated dataset.
  • Age and pose category bins = not specified
    The balancing procedure requires defining age and pose categories, but the paper does not specify the binning scheme in the main text. This is a free design choice that could affect the resulting dataset diversity.
assumptions (4)
  • domain assumption FairFace attribute labels are sufficiently accurate for balancing and evaluation
    The pipeline uses FairFace for ethnicity, gender, and age inference. Supplementary Table 4 reports per-group accuracy as low as 0.581, so the assumption is fragile.
  • standard math Linearity of log-odds in the logit model
    Equation (2) assumes a linear relationship between log-odds and attributes. The authors provide DHARMa diagnostics in supplementary D, but this remains a modeling assumption.
  • domain assumption The considered attributes (gender, ethnicity, age, pose) are the main sources of bias in face verification
    The paper's entire intervention is to balance these four attributes, implicitly assuming other attributes (e.g., hairstyle, makeup, image quality) are less important.
  • domain assumption The DCFace pipeline produces usable images when conditions are matched
    The generation relies on the DCFace pipeline, and the authors had to add segment matching to ensure convergence, indicating the base pipeline is fragile.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Fairer Analysis and Demographically Balanced Face Generation for Fairer Face Verification." pith.science (2026). https://pith.science/paper/DQHQ5KZ6

@misc{pith2026241203349,
  author       = {Pith},
  title        = {Pith review of: Fairer Analysis and Demographically Balanced Face Generation for Fairer Face Verification},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DQHQ5KZ6}},
  note         = {Machine review of arXiv:2412.03349}
}
read the original abstract

Face recognition and verification are two computer vision tasks whose performances have advanced with the introduction of deep representations. However, ethical, legal, and technical challenges due to the sensitive nature of face data and biases in real-world training datasets hinder their development. Generative AI addresses privacy by creating fictitious identities, but fairness problems remain. Using the existing DCFace SOTA framework, we introduce a new controlled generation pipeline that improves fairness. Through classical fairness metrics and a proposed in-depth statistical analysis based on logit models and ANOVA, we show that our generation pipeline improves fairness more than other bias mitigation approaches while slightly improving raw performance.

Figures

Figures reproduced from arXiv: 2412.03349 by the authors.

Figure 1
Figure 1. Comparison of the face verification fairness (equalized [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Global pipeline overview for training and evaluating models with the baselines and our proposed generative approach. Critical [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Detailed view of our controlled generation method. [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (14 more)
Figure 4
Figure 4. Figure 4: Attribute analysis of the evaluation datasets. Attributes [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: Marginal effect on FMR (lower is better) for each method compared to the unprotected group. Example: ”When using CASIA , [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: ANOVA results on RFW: total height corresponds to [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: QQ-plot of residuals and Residual vs. predicted plot: [PITH_FULL_IMAGE:figures/full_fig_p014_7.png]
Figure 8
Figure 8. Figure 8: Residual vs. predictor plots: exogeneity is verified. [PITH_FULL_IMAGE:figures/full_fig_p015_8.png]
Figure 9
Figure 9. Figure 9: Overdispersion Test: Correct Specification and no auto [PITH_FULL_IMAGE:figures/full_fig_p015_9.png]
Figure 10
Figure 10. Figure 10: Zero-inflation Test: the model correctly predicts the [PITH_FULL_IMAGE:figures/full_fig_p015_10.png]
Figure 11
Figure 11. Figure 11: Marginal effect on FMR (lower is better) for each method compared to the unprotected group. Analysis done on FAVCI2D [PITH_FULL_IMAGE:figures/full_fig_p016_11.png]
Figure 12
Figure 12. Figure 12: Marginal effect on TMR (lower in absolute is better) for each method compared to the unprotected group. Analysis done on [PITH_FULL_IMAGE:figures/full_fig_p016_12.png]
Figure 13
Figure 13. Figure 13: ANOVA results on FAVCI2D : total height corresponds to [PITH_FULL_IMAGE:figures/full_fig_p017_13.png]
Figure 14
Figure 14. Figure 14: Marginal effect on FMR (lower is better) for each method compared to the unprotected group. Analysis done on BFW [PITH_FULL_IMAGE:figures/full_fig_p017_14.png]
Figure 15
Figure 15. Figure 15: Marginal effect on TMR (lower in absolute is better) for each method compared to the unprotected group. Analysis done on [PITH_FULL_IMAGE:figures/full_fig_p017_15.png]
Figure 16
Figure 16. Figure 16: ANOVA results on BFW: total height corresponds to [PITH_FULL_IMAGE:figures/full_fig_p018_16.png]
Figure 17
Figure 17. Figure 17: Marginal effects on TMR (lower in absolute is better) for each method compared to the unprotected group. Analysis done on [PITH_FULL_IMAGE:figures/full_fig_p018_17.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

89 extracted references · 69 canonical work pages

  1. [1]

    A reductions approach to fair classification

    Alekh Agarwal, Alina Beygelzimer, Miroslav Dudik, John Langford, and Hanna Wallach. A reductions approach to fair classification. In Jennifer Dy and Andreas Krause, ed- itors, Proceedings of the 35th International Conference on Machine Learning, volume 80 of Proceedings of Machine Learning Research, pages 60–69. PMLR, 10–15 Jul 2018. 5

    2018

  2. [2]

    Fair regression: Quantitative definitions and reduction-based algorithms

    Alekh Agarwal, Miroslav Dudik, and Zhiwei Steven Wu. Fair regression: Quantitative definitions and reduction-based algorithms. In Kamalika Chaudhuri and Ruslan Salakhut- dinov, editors, Proceedings of the 36th International Con- ference on Machine Learning, volume 97 of Proceedings of Machine Learning Research, pages 120–129. PMLR, 09–15 Jun 2019. 5

    2019

  3. [3]

    Is face recognition sexist? no, gendered hairstyles and biology are, 08 2020

    V ´ıtor Albiero and Kevin Bowyer. Is face recognition sexist? no, gendered hairstyles and biology are, 08 2020. 3

    2020

  4. [4]

    How does gender balance in training data affect face recognition accu- racy? In 2020 ieee international joint conference on biomet- rics (ijcb), pages 1–10

    V ´ıtor Albiero, Kai Zhang, and Kevin W Bowyer. How does gender balance in training data affect face recognition accu- racy? In 2020 ieee international joint conference on biomet- rics (ijcb), pages 1–10. IEEE, 2020. 2

    2020

  5. [5]

    Gendered differences in face recognition accu- racy explained by hairstyles, makeup, and facial morphol- ogy

    V ´ıtor Albiero, Kai Zhang, Michael C King, and Kevin W Bowyer. Gendered differences in face recognition accu- racy explained by hairstyles, makeup, and facial morphol- ogy. IEEE Transactions on Information Forensics and Secu- rity, 17:127–137, 2021. 2

    2021

  6. [6]

    Adnan Amin, Sajid Anwar, Awais Adnan, Muhammad Nawaz, Newton Howard, Junaid Qadir, Ahmad Y . A. Hawalah, and Amir Hussain. Comparing oversampling tech- niques to handle the class imbalance problem: A customer churn prediction case study. IEEE Access , 4:7940–7957,

  7. [7]

    Mostly harmless econometrics: An empiricist’s companion

    Joshua D Angrist and J ¨orn-Steffen Pischke. Mostly harmless econometrics: An empiricist’s companion. Princeton univer- sity press, 2009. 5

    2009

  8. [8]

    Digiface-1m: 1 million digi- tal face images for face recognition

    Gwangbin Bae, Martin de La Gorce, Tadas Baltru ˇsaitis, Charlie Hewitt, Dong Chen, Julien Valentin, Roberto Cipolla, and Jingjing Shen. Digiface-1m: 1 million digi- tal face images for face recognition. In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pages 3526–3535, 2023. 1, 2, 3, 4, 6, 13, 19

    2023

Show all 89 references
  1. [9]

    Attribute-preserving face dataset anonymization via latent code optimization

    Simone Barattin, Christos Tzelepis, Ioannis Patras, and Nicu Sebe. Attribute-preserving face dataset anonymization via latent code optimization. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 8001–8010, 2023. 2

    2023

  2. [10]

    Mahakil: Diversity based oversampling approach to alleviate the class imbalance issue in software defect prediction

    Kwabena Ebo Bennin, Jacky Wai Keung, Passakorn Phan- nachitta, Akito Monden, and Solomon Mensah. Mahakil: Diversity based oversampling approach to alleviate the class imbalance issue in software defect prediction. IEEE Trans- actions on Software Engineering, 44:534–550, 2018. 4, 13

    2018

  3. [11]

    Glasses Detector, 3 2024

    Mantas Bir ˇskus. Glasses Detector, 3 2024. 13

    2024

  4. [12]

    Idiff-face: Synthetic-based face recognition through fizzy identity-conditioned diffusion model

    Fadi Boutros, Jonas Henry Grebe, Arjan Kuijper, and Naser Damer. Idiff-face: Synthetic-based face recognition through fizzy identity-conditioned diffusion model. In Proceedings of the IEEE/CVF International Conference on Computer Vi- sion, pages 19650–19661, 2023. 8

    2023

  5. [13]

    Safe-level-smote: Safe-level- synthetic minority over-sampling technique for handling the class imbalanced problem

    Chumphol Bunkhumpornpat, Krung Sinapiromsaran, and Chidchanok Lursinsap. Safe-level-smote: Safe-level- synthetic minority over-sampling technique for handling the class imbalanced problem. In Thanaruk Theeramunkong, Boonserm Kijsirikul, Nick Cercone, and Tu-Bao Ho, editors, Ad...

    2009

  6. [14]

    Gender shades: Inter- sectional accuracy disparities in commercial gender classifi- cation

    Joy Buolamwini and Timnit Gebru. Gender shades: Inter- sectional accuracy disparities in commercial gender classifi- cation. In Conference on fairness, accountability and trans- parency, pages 77–91, 2018. 1, 2

    2018

  7. [15]

    Vggface2: A dataset for recognising faces across pose and age

    Qiong Cao, Li Shen, Weidi Xie, Omkar M Parkhi, and Andrew Zisserman. Vggface2: A dataset for recognising faces across pose and age. In 2018 13th IEEE International Conference on Automatic Face & Gesture Recognition (FG 2018), pages 67–74. IEEE, 2018. 1, 2

    2018

  8. [16]

    Kegelmeyer

    Nitesh Chawla, Kevin Bowyer, Lawrence Hall, and W. Kegelmeyer. Smote: Synthetic minority over-sampling tech- nique. J. Artif. Intell. Res. (JAIR), 16:321–357, 06 2002. 13

    2002

  9. [17]

    Mitigating gender bias in face recognition using the von mises-fisher mixture model

    Jean-R ´emy Conti, Nathan Noiry, Stephan Clemencon, Vin- cent Despiegel, and St ´ephane Gentric. Mitigating gender bias in face recognition using the von mises-fisher mixture model. In International Conference on Machine Learning , pages 4344–4369. PMLR, 2022. 3, 4

    2022

  10. [18]

    Vera-Rodr´ıguez, Minchul Kim, C

    Ivan Deandres-Tame, Rub ´en Tolosana, Pietro Melzi, R. Vera-Rodr´ıguez, Minchul Kim, C. Rathgeb, Xiaoming Liu, A. Morales, Julian Fi´errez, J. Ortega-Garcia, Zhizhou Zhong, Y . Huang, Yuxi Mi, Shouhong Ding, Shuigeng Zhou, Shuai He, Lingzhi Fu, Heng Cong, Rongyu Zhang, Zhihong...

    2024

  11. [19]

    Improving fairness using vision-language driven im- age augmentation

    Moreno D’Inc `a, Christos Tzelepis, Ioannis Patras, and Nicu Sebe. Improving fairness using vision-language driven im- age augmentation. In Proceedings of the IEEE/CVF Win- ter Conference on Applications of Computer Vision , pages 4695–4704, 2024. 1

    2024

  12. [20]

    Multi-attribute balanced sampling for disentangled GAN controls

    Perla Doubinsky, Nicolas Audebert, Michel Crucianu, and Herv´e Le Borgne. Multi-attribute balanced sampling for disentangled GAN controls. Pattern Recognition Letters , 162:56–62, 2022. 2

    2022

  13. [21]

    Wasserstein loss for semantic editing in the latent space of gans

    Perla Doubinsky, Nicolas Audebert, Michel Crucianu, and Herv´e Le Borgne. Wasserstein loss for semantic editing in the latent space of gans. In International Conference on Content-Based Multimedia Indexing, 2023. 2

    2023

  14. [22]

    Domain shift and emerging questions in facial recogni- tion technology

    Ho Daniel E., Emily Black, Maneesh Agrawala, and Fei-Fei Li. Domain shift and emerging questions in facial recogni- tion technology. HAI Policy Brief, 2020. 1, 2, 4

    2020

  15. [23]

    Dharma: Residual diagnostics for hierarchical (multi-level / mixed) regression models., 2018

    Hartig F. Dharma: Residual diagnostics for hierarchical (multi-level / mixed) regression models., 2018. 14

    2018

  16. [24]

    A survey on bias in visual datasets

    Simone Fabbrizzi, Symeon Papadopoulos, Eirini Ntoutsi, and Ioannis Kompatsiaris. A survey on bias in visual datasets. Computer Vision and Image Understanding , 223:103552, 2022. 2

    2022

  17. [25]

    When do gans replicate? on the choice of dataset size

    Qianli Feng, Chenqi Guo, Fabian Benitez-Quiroz, and Aleix M Martinez. When do gans replicate? on the choice of dataset size. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 6701–6710, 2021. 2

    2021

  18. [26]

    Ruwani M

    K. Ruwani M. Fernando and Chris P. Tsokos. Dynami- cally weighted balanced loss: Class imbalanced learning and confidence calibration of deep neural networks. IEEE Transactions on Neural Networks and Learning Systems , 33(7):2940–2951, 2022. 4, 14

    2022

  19. [27]

    An introduction to statistical learning: with appli- cations in R

    James Gareth, Witten Daniela, Hastie Trevor, and Tibshirani Robert. An introduction to statistical learning: with appli- cations in R. Spinger, 2013. 5

    2013

  20. [28]

    Jointly de-biasing face recognition and demographic attribute estimation

    Sixue Gong, Xiaoming Liu, and A Jain. Jointly de-biasing face recognition and demographic attribute estimation. In ECCV, pages 330–347, 2020. 5

    2020

  21. [29]

    Face recognition vendor test part 3: demographic effects

    Patrick J Grother, Mei L Ngan, Kayee K Hanaoka, et al. Face recognition vendor test part 3: demographic effects. Techni- cal report, National Institute of Standards and Technology,

  22. [30]

    Ms-celeb-1m: A dataset and benchmark for large-scale face recognition

    Yandong Guo, Lei Zhang, Yuxiao Hu, Xiaodong He, and Jianfeng Gao. Ms-celeb-1m: A dataset and benchmark for large-scale face recognition. In European conference on computer vision, pages 87–102. Springer, 2016. 1, 2

    2016

  23. [31]

    Borderline-smote: A new over-sampling method in imbalanced data sets learning

    Hui Han, Wen-Yuan Wang, and Bing-Huan Mao. Borderline-smote: A new over-sampling method in imbalanced data sets learning. In De-Shuang Huang, Xiao- Ping Zhang, and Guang-Bin Huang, editors, Advances in Intelligent Computing, pages 878–887, Berlin, Heidelberg,

  24. [32]

    Equal- ity of opportunity in supervised learning

    Moritz Hardt, Eric Price, Eric Price, and Nati Srebro. Equal- ity of opportunity in supervised learning. In D. Lee, M. Sugiyama, U. Luxburg, I. Guyon, and R. Garnett, editors, Advances in Neural Information Processing Systems , vol- ume 29. Curran Associates, Inc., 2016. 5

    2016

  25. [33]

    Regression with dummy variables

    Melissa A Hardy. Regression with dummy variables. Num- ber 93. Sage, 1993. 6

    1993

  26. [34]

    Harvey, Adam

    Jules. Harvey, Adam. LaPlace. Exposing.ai, 2021. 1

    2021

  27. [35]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proceed- ings of the IEEE conference on computer vision and pattern recognition, pages 770–778, 2016. 3

    2016

  28. [36]

    6d rotation representation for unconstrained head pose estimation

    Thorsten Hempel, Ahmed A Abdelrahman, and Ayoub Al- Hamadi. 6d rotation representation for unconstrained head pose estimation. In 2022 IEEE International Conference on Image Processing (ICIP), pages 2496–2500. IEEE, 2022. 3

    2022

  29. [37]

    Denoising dif- fusion probabilistic models

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising dif- fusion probabilistic models. In H. Larochelle, M. Ranzato, R. Hadsell, M.F. Balcan, and H. Lin, editors, Advances in Neural Information Processing Systems , volume 33, pages 6840–6851. Curran Associates, Inc., 2020. 3

    2020

  30. [38]

    Learning deep representation for imbalanced classi- fication

    Chen Huang, Yining Li, Chen Change Loy, and Xiaoou Tang. Learning deep representation for imbalanced classi- fication. In 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 5375–5384, 2016. 4, 14

    2016

  31. [39]

    Huang and Erik Learned-Miller

    Gary B. Huang and Erik Learned-Miller. Labeled faces in the wild: Updates and new reporting procedures. Techni- cal Report UM-CS-2014-003, University of Massachusetts, Amherst, May 2014. 4, 8, 13

    2014

  32. [40]

    Hupont and Carles Fern ´andez Tena

    I. Hupont and Carles Fern ´andez Tena. Demogpairs: Quan- tifying the impact of demographic imbalance in deep face recognition. IEEE International Conference on Automatic Face & Gesture Recognition, 2019. 15

    2019

  33. [41]

    Simple data balancing achieves com- petitive worst-group-accuracy

    Badr Youbi Idrissi, Martin Arjovsky, Mohammad Pezeshki, and David Lopez-Paz. Simple data balancing achieves com- petitive worst-group-accuracy. In Bernhard Sch ¨olkopf, Car- oline Uhler, and Kun Zhang, editors,Proceedings of the First Conference on Causal Learning and Reasonin...

  34. [42]

    Fairface: Face at- tribute dataset for balanced race, gender, and age

    Kimmo K ¨arkk¨ainen and Jungseock Joo. Fairface: Face at- tribute dataset for balanced race, gender, and age. arXiv preprint arXiv:1908.04913, 2019. 1, 2, 3, 5, 8

    1908 arXiv

  35. [43]

    The megaface benchmark: 1 million faces for recognition at scale

    Ira Kemelmacher-Shlizerman, Steven M Seitz, Daniel Miller, and Evan Brossard. The megaface benchmark: 1 million faces for recognition at scale. In Proceedings of the IEEE conference on computer vision and pattern recogni- tion, pages 4873–4882, 2016. 1, 2

    2016

  36. [44]

    Adaface: Quality adaptive margin for face recognition

    Minchul Kim, Anil K Jain, and Xiaoming Liu. Adaface: Quality adaptive margin for face recognition. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 18750–18759, 2022. 3, 13, 14

    2022

  37. [45]

    Dc- face: Synthetic face generation with dual condition diffu- sion model

    Minchul Kim, Feng Liu, Anil Jain, and Xiaoming Liu. Dc- face: Synthetic face generation with dual condition diffu- sion model. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 12715– 12725, 2023. 1, 2, 3, 4, 6, 8, 13, 19

    2023

  38. [46]

    Over- sampling for imbalanced learning based on k-means and smote

    Felix Last, Georgios Douzas, and Fernando Bac ¸ ˜ao. Over- sampling for imbalanced learning based on k-means and smote. ArXiv, abs/1711.00837, 2017. 4, 13

    2017 arXiv

  39. [47]

    Subclass-based under- sampling for class-imbalanced image classification

    Daniel Lehmann and Marc Ebner. Subclass-based under- sampling for class-imbalanced image classification. In VISI- GRAPP, 2022. 4, 14

    2022

  40. [48]

    Exploratory under-sampling for class-imbalance learning

    Xu-ying Liu, Jianxin Wu, and Zhi-hua Zhou. Exploratory under-sampling for class-imbalance learning. In Sixth In- ternational Conference on Data Mining (ICDM’06) , pages 965–969, 2006. 4, 14

    2006

  41. [49]

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

    Stylianos Moschoglou, Athanasios Papaioannou, Chris- tos Sagonas, Jiankang Deng, Irene Kotsia, and Stefanos Zafeiriou. Agedb: the first manually collected, in-the-wild age database. In proceedings of the IEEE conference on computer vision and pattern recognition workshops , pa...

    2017

  42. [50]

    Neto, Eduarda Caldeira, Jaime S

    Pedro C. Neto, Eduarda Caldeira, Jaime S. Cardoso, and Ana F. Sequeira. Compressed models decompress race bi- ases: What quantized models forget for fair face recognition. 2023 International Conference of the Biometrics Special In- terest Group (BIOSIG), pages 1–5, 2023. 1, 2

    2023

  43. [51]

    Analyzing bias in diffusion-based face generation models

    Malsha V Perera and Vishal M Patel. Analyzing bias in diffusion-based face generation models. arXiv preprint arXiv:2305.06402, 2023. 1, 2

    2023 arXiv

  44. [52]

    The good, the bad, and the ugly face challenge problem

    P Jonathon Phillips, J Ross Beveridge, Bruce A Draper, Geof Givens, Alice J O’Toole, David Bolme, Joseph Dun- lop, Yui Man Lui, Hassan Sahibzada, and Samuel Weimer. The good, the bad, and the ugly face challenge problem. Im- age and Vision Computing, 30(3):177–185, 2012. 2

    2012

  45. [53]

    Face verification with chal- lenging imposters and diversified demographics

    Adrian Popescu, Liviu-Daniel S , tefan, J ´erˆome Deshayes- Chossart, and Bogdan Ionescu. Face verification with chal- lenging imposters and diversified demographics. InProceed- ings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pages 3357–3366, 2022. 2...

    2022

  46. [54]

    Synface: Face recognition with syn- thetic data

    Haibo Qiu, Baosheng Yu, Dihong Gong, Zhifeng Li, Wei Liu, and Dacheng Tao. Synface: Face recognition with syn- thetic data. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 10880–10890, 2021. 1, 2, 3, 4, 13, 19

    2021

  47. [55]

    Does training ai violate copyright law? Berke- ley Tech

    Jenny Quang. Does training ai violate copyright law? Berke- ley Tech. LJ, 36:1407, 2021. 2

    2021

  48. [56]

    About face: A survey of facial recognition evaluation, 2021

    Inioluwa Deborah Raji and Genevieve Fried. About face: A survey of facial recognition evaluation, 2021. 4

    2021

  49. [57]

    Face recognition: too bias, or not too bias? In Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition Work- shops, pages 0–1, 2020

    Joseph P Robinson, Gennady Livitz, Yann Henon, Can Qin, Yun Fu, and Samson Timoner. Face recognition: too bias, or not too bias? In Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition Work- shops, pages 0–1, 2020. 1, 2, 3, 4, 6

    2020

  50. [58]

    Robinson, Can Qin, Yann Henon, Samson Timo- ner, and Yun Fu

    Joseph P. Robinson, Can Qin, Yann Henon, Samson Timo- ner, and Yun Fu. Balancing biases and preserving privacy on balanced faces in the wild. IEEE Transactions on Image Processing, 32:4365–4377, 2023. 3, 15

    2023

  51. [59]

    Flac: Fairness-aware representation learning by suppressing attribute-class associations

    Ioannis Sarridis, Christos Koutlis, Symeon Papadopoulos, and Christos Diou. Flac: Fairness-aware representation learning by suppressing attribute-class associations. arXiv preprint arXiv:2304.14252, 2023. 2

    2023 arXiv

  52. [60]

    Towards fair face verification: An in-depth analysis of demographic biases

    Ioannis Sarridis, Christos Koutlis, Symeon Papadopoulos, and Christos Diou. Towards fair face verification: An in-depth analysis of demographic biases. arXiv preprint arXiv:2307.10011, 2023. 1, 2

    2023 arXiv

  53. [61]

    Towards fair face verification: An in- depth analysis of demographic biases, 2023

    Ioannis Sarridis, Christos Koutlis, Symeon Papadopoulos, and Christos Diou. Towards fair face verification: An in- depth analysis of demographic biases, 2023. 3, 8

    2023

  54. [62]

    Towards fair face verification: An in- depth analysis of demographic biases, 2023

    Ioannis Sarridis, Christos Koutlis, Symeon Papadopoulos, and Christos Diou. Towards fair face verification: An in- depth analysis of demographic biases, 2023. 8

    2023

  55. [63]

    Bloom: A 176b-parameter open-access multilingual language model

    Teven Le Scao, Angela Fan, Christopher Akiki, Ellie Pavlick, Suzana Ili ´c, Daniel Hesslow, Roman Castagn ´e, Alexandra Sasha Luccioni, Franc ¸ois Yvon, Matthias Gall´e, et al. Bloom: A 176b-parameter open-access multilingual language model. arXiv preprint arXiv:2211.05100, 2022. 2

    2022 arXiv

  56. [64]

    Facenet: A unified embedding for face recognition and clus- tering

    Florian Schroff, Dmitry Kalenichenko, and James Philbin. Facenet: A unified embedding for face recognition and clus- tering. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 815–823, 2015. 2

    2015

  57. [65]

    Neil Selwyn, Mark Andrejevic, Gavin J. D. Smith, Xin Gu, and Christopher O’Neill. Facial recognition technology: key issues and emerging concerns, 1 2023. 1

    2023

  58. [66]

    Frontal to profile face verification in the wild

    Soumyadip Sengupta, Jun-Cheng Chen, Carlos Castillo, Vishal M Patel, Rama Chellappa, and David W Jacobs. Frontal to profile face verification in the wild. In 2016 IEEE winter conference on applications of computer vision (WACV), pages 1–9. IEEE, 2016. 13

    2016

  59. [67]

    Diffusion art or digital forgery? investigating data replication in diffusion models

    Gowthami Somepalli, Vasu Singla, Micah Goldblum, Jonas Geiping, and Tom Goldstein. Diffusion art or digital forgery? investigating data replication in diffusion models. In Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 6048–6058, 2023. 2

    2023

  60. [68]

    Hassan, and Ken ichi Matsumoto

    Chakkrit Kla Tantithamthavorn, A. Hassan, and Ken ichi Matsumoto. The impact of class rebalancing techniques on the performance and interpretation of defect prediction mod- els. IEEE Transactions on Software Engineering, 46:1200– 1219, 2018. 13

    2018

  61. [69]

    A comprehensive study on face recognition biases beyond demographics

    Philipp Terh ¨orst, Jan Niklas Kolf, Marco Huber, Florian Kirchbuchner, Naser Damer, Aythami Morales, Julian Fier- rez, and Arjan Kuijper. A comprehensive study on face recognition biases beyond demographics. arXiv preprint arXiv:2103.01592, 2021. 2

    2021 arXiv

  62. [70]

    Comparison-level mitigation of ethnic bias in face recognition

    Philipp Terh ¨orst, Mai Ly Tran, Naser Damer, Florian Kirch- buchner, and Arjan Kuijper. Comparison-level mitigation of ethnic bias in face recognition. In 2020 8th international workshop on biometrics and forensics (IWBF) , pages 1–6. IEEE, 2020. 2

    2020

  63. [71]

    Under-sampling class imbalanced datasets by combin- ing clustering analysis and instance selection

    Chih-Fong Tsai, Wei-Chao Lin, Ya-Han Hu, and Guan-Ting Yao. Under-sampling class imbalanced datasets by combin- ing clustering analysis and instance selection. Information Sciences, 477:47–54, 2019. 4, 14

    2019

  64. [72]

    The ethical questions that haunt facial-recognition research

    Richard Van Noorden. The ethical questions that haunt facial-recognition research. Nature, 587(7834):354–358,

  65. [73]

    Mitigating bias in face recog- nition using skewness-aware reinforcement learning

    Mei Wang and Weihong Deng. Mitigating bias in face recog- nition using skewness-aware reinforcement learning. InPro- ceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 9322–9331, 2020. 2

    2020

  66. [74]

    Racial faces in the wild: Reducing racial bias by information maximization adaptation network

    Mei Wang, Weihong Deng, Jiani Hu, Xunqiang Tao, and Yaohai Huang. Racial faces in the wild: Reducing racial bias by information maximization adaptation network. In Proceedings of the IEEE International Conference on Com- puter Vision, pages 692–702, 2019. 1, 2, 6

    2019

  67. [75]

    Meta bal- anced network for fair face recognition

    Mei Wang, Yaobin Zhang, and Weihong Deng. Meta bal- anced network for fair face recognition. IEEE transactions on pattern analysis and machine intelligence , 44(11):8433– 8448, 2021. 2, 4, 8, 20

    2021

  68. [76]

    Learn- ing to model the tail

    Yu-Xiong Wang, Deva Ramanan, and Martial Hebert. Learn- ing to model the tail. In Proceedings of the 31st Interna- tional Conference on Neural Information Processing Sys- tems, NIPS’17, page 7032–7042, Red Hook, NY , USA,

  69. [77]

    Fake it till you make it: face analysis in the wild using synthetic data alone

    Erroll Wood, Tadas Baltru ˇsaitis, Charlie Hewitt, Sebastian Dziadzio, Thomas J Cashman, and Jamie Shotton. Fake it till you make it: face analysis in the wild using synthetic data alone. In Proceedings of the IEEE/CVF international conference on computer vision, pages 3681–36...

    2021

  70. [78]

    Ramface: Race adaptive mar- gin based face recognition for racial bias mitigation

    Zhanjia Yang, Xiangping Zhu, Changyuan Jiang, Wen- shuang Liu, and Linlin Shen. Ramface: Race adaptive mar- gin based face recognition for racial bias mitigation. In 2021 IEEE International Joint Conference on Biometrics (IJCB) , pages 1–8. IEEE, 2021. 1, 2

    2021

  71. [79]

    Learn- ing face representation from scratch

    Dong Yi, Zhen Lei, Shengcai Liao, and Stan Z Li. Learn- ing face representation from scratch. arXiv preprint arXiv:1411.7923, 2014. 2, 3, 4, 20

    2014 arXiv

  72. [80]

    Seyma Yucer, Furkan Tektas, Noura Al Moubayed, and Toby P. Breckon. Measuring hidden bias within face recog- nition via racial phenotypes. In2022 IEEE/CVF Winter Con- ference on Applications of Computer Vision (WACV), pages 3202–3211, 2022. 3

    2022

  73. [81]

    Towards pose invariant face recogni- tion in the wild

    Jian Zhao, Yu Cheng, Yan Xu, Lin Xiong, Jianshu Li, Fang Zhao, Karlekar Jayashree, Sugiri Pranata, Shengmei Shen, Junliang Xing, et al. Towards pose invariant face recogni- tion in the wild. In Proceedings of the IEEE conference on computer vision and pattern recognition, page...

  74. [82]

    Dual-agent gans for photorealistic and identity preserving profile face synthesis

    Jian Zhao, Lin Xiong, Panasonic Karlekar Jayashree, Jian- shu Li, Fang Zhao, Zhecan Wang, Panasonic Sugiri Pranata, Panasonic Shengmei Shen, Shuicheng Yan, and Jiashi Feng. Dual-agent gans for photorealistic and identity preserving profile face synthesis. Advances in neural in...

    2017

  75. [83]

    Cross-pose lfw: A database for studying cross-pose face recognition in un- constrained environments

    Tianyue Zheng and Weihong Deng. Cross-pose lfw: A database for studying cross-pose face recognition in un- constrained environments. Beijing University of Posts and Telecommunications, Tech. Rep, 5(7), 2018. 4, 13

    2018

  76. [84]

    Cross-age lfw: A database for studying cross-age face recognition in un- constrained environments

    Tianyue Zheng, Weihong Deng, and Jiani Hu. Cross-age lfw: A database for studying cross-age face recognition in un- constrained environments. arXiv preprint arXiv:1708.08197,

    arXiv

  77. [85]

    Oversampling method for imbalanced classification

    Zhuoyuan Zheng, Yunpeng Cai, and Ye Li. Oversampling method for imbalanced classification. Comput. Informatics, 34:1017–1037, 2015. 4, 13

    2015

  78. [86]

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

    Zheng Zhu et al. Webface260m: A benchmark unveiling the power of million-scale deep face recognition. In IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2021. 13 Supplementary Material A. Parameters for training and generation For training the face classifie...

    2021

  79. [351]

    PMLR, 11–13 Apr 2022. 13

    2022

  80. [2005]

    Springer Berlin Heidelberg. 13

  81. [2017]

    Curran Associates Inc. 4, 14

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.