FaceX: Understanding Face Attribute Classifiers through Summary Model Explanations

Christos Diou; Christos Koutlis; Ioannis Sarridis; Symeon Papadopoulos

REVIEW 4 major objections 5 minor 45 references

FaceX: Understanding Face Attribute Classifiers through Summary Model Explanations

T0 review · 4 major / 5 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 FaceX claims to be the first summary model explanation method for face attribute classifiers, aggregating per-image attributions across 19 facial regions to expose attribute shortcuts.

desk verdict FaceX is a useful summary-explanation tool for face classifiers, but its 'high effectiveness' claim outruns an evaluation built on an unvalidated ranking premise. read the letter →

arxiv 2412.07313 v1 pith:GV7IX4LO submitted 2024-12-10 cs.CV

Ioannis Sarridis , Christos Koutlis , Symeon Papadopoulos , Christos Diou This is my paper · ORCID

classification cs.CV

keywords explainableAIfaceattributeclassificationsummarymodelexplanationsGrad-CAMbiasdetectionfacialregionparsingIntersectionoverfairness

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

FaceX introduces a way to summarize how a face attribute classifier behaves across a whole test set, instead of leaving a human to stare at thousands of individual heatmaps. It aggregates Grad-CAM attributions over 19 predefined facial regions into a single IoR score per region, then visualizes the ranking on an abstract face and shows the highest-impact image patches for each region. The paper claims that when a model exploits a spurious correlation, such as Gender with Wearing_Lipstick, the corresponding region rises to the top of the IoR ranking, exposing the shortcut. Across controlled and real bias scenarios on CelebA, FairFace, CelebAMask-HQ, and RFW, it reports mean ranking positions of 2.25 for single-attribute biases and 1.25 for multi-attribute biases, and shows a visible shift in the summary after applying a bias mitigation method. A reader should care because this is a candidate mechanism for auditing facial analysis models at the model level, objectively and repeatably.

What carries the argument

The load-bearing object is the Intersection over Region (IoR) score: for sample $i$ and region $r$, $\mathrm{IoR}_{i,r} = \sum_{h,w}(G_i \odot M_{i,r})_{h,w} / \sum_{h,w}(M_{i,r})_{h,w}$, the average Grad-CAM attribution inside the region's binary mask, and $\mathrm{IoR}_r$ is its mean over test samples containing that region. Ranking the 19 IoR values on a face prototype converts many individual heatmaps into one comparable picture of where the model looks. To answer 'what' triggers that focus, FaceX additionally computes patch-level attribution $V_{i,q,r} = \sum_{h,w}(G_i \odot M_{i,r} \odot P'_{i,q})_{h,w}$ and selects the $k$ highest-scoring patches per region. Face-parsing masks come from CelebAMask-HQ annotations or, when absent, from the FaRL face parser; Grad-CAM supplies the underlying instance-level attribution.

What would settle it

A concrete check: train a model on the 99% Gender-Wearing_Eyeglasses correlation, then compare IoR ranking with occlusion sensitivity by masking the eyeglasses region and measuring prediction change. If occlusion shows eyeglasses are the decisive shortcut while FaceX places them 6th (as in Table 1), the region-ranking method fails to expose the true bias; if IoR rank tracks occlusion sensitivity across all seven attributes, the method's core claim is corroborated.

Watch

Extended reading notes

Core claim

On its own terms, the paper's discovery is that instance-level explanations, which are noisy and inconsistent across individual images, can be lifted to a stable model-level summary by exploiting the fixed geometry of faces. The key output is a ranked list of 19 regions, computed as average Intersection over Region between Grad-CAM heatmaps and face-parsing masks, together with the top-k patches that carry the strongest activation inside each region. The paper asserts that this summary reliably identifies both single-attribute shortcuts (e.g., lipstick for gender, smiling for age) and, to a lesser degree, multi-attribute shortcuts, and that it can reveal biases a model inherits from unbalanced training data even when no bias was injected. It also demonstrates that applying the FLAC mitigation procedure changes the summary in the expected direction, with hair and brow regions dropping out of the top of the ranking.

Load-bearing premise

The evaluation assumes a model biased toward an attribute will concentrate its Grad-CAM activations in the corresponding facial region, making that region's IoR rank high; this premise is not guaranteed, and it fails for eyeglasses and earrings in Table 1.

Editorial extensions

If this is right

Practitioners can audit a face attribute classifier with a single heatmap and a patch grid rather than manually inspecting large numbers of per-image explanations.
Spurious correlations such as Gender-Lipstick, Age-Smiling, and Gender-Blond_Hair should appear as top-ranked IoR regions, making shortcut features easy to localize.
Bias mitigation methods such as FLAC change the IoR distribution in measurable ways, so FaceX can serve as a before-and-after diagnostic for fairness interventions.
When region annotations are unavailable, a face parser can supply masks, allowing the same summary to be produced for arbitrary test benchmarks such as RFW.
In multi-attribute bias settings, FaceX typically places at least one biased region first, but only about a third of its experiments place both biased regions in the top three, reflecting models' tendency to lean on the easiest shortcut.

Reading between the lines

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

Because FaceX is a summarization layer over whatever attribution map is plugged in, its region rankings may change if Grad-CAM is replaced by another explainer; testing the same bias scenarios with a different attribution method would show how much of the result is method-dependent.
The ranking-position metric assumes a biased model concentrates activation on the biased region, and the eyeglasses (rank 6) and earrings (rank 5) cases in Table 1 show this assumption can fail; a stronger test would compare IoR rankings against occlusion-based measurement of true shortcut reliance.
The same region-aggregation idea transfers naturally to any vision domain with fixed part geometry, such as fine-grained vehicle or animal classification, where a part parser and an attribution method would suffice to produce the summary.
A testable extension is to feed FaceX summaries back into training, penalizing high IoR on regions tied to protected attributes; the paper lists this as future work, but the mechanism is directly available from the method's outputs.

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.

Desk Editor's Note

read the letter

Colleague,

You should know two things about FaceX. First, the core idea is genuinely useful: instead of asking a human to stare at hundreds of Grad-CAM maps, the authors parse each face into 19 regions, average the per-region activation over a test set, and render one heatmap on a canonical face plus the top-k contributing patches. That is a clean, sensible way to get a summary read on where a face attribute classifier looks. The code is available, the method is described precisely, and the paper is honest and readable.

Second, the evaluation does not back the 'high effectiveness' claim. The quantitative metric is the rank of the biased region in the IoR ordering. That metric assumes a biased model will concentrate its activation in the region tied to the shortcut. The authors actually say in §4.2 that this premise 'might not align with the behavior of deep learning models'—and then Table 1 shows it failing: eyeglasses rank 6th, earrings 5th. The mean rank of 2.25 hides those misses. Without a control condition with no injected bias, a permutation baseline, or error bars over model initializations, we cannot tell whether IoR ranks track actual decision-relevant features or just large/diffuse regions. The high-impact patch selection reuses the same IoR values, so it cannot independently confirm the rankings.

That is the load-bearing soft spot. There are smaller ones: no comparison to SP-LIME, CRP, or other summary-explanation methods; patch size Z and the Grad-CAM layer are not specified; the multi-attribute results show the second biased region often ranking 6th–12th, which the authors explain away plausibly but do not test.

What the paper does well: the aggregation itself is novel enough for this domain, the face-prototype visualization is an effective communication device, and the RFW and FLAC experiments show the method can produce sensible, practically relevant output. The honest limitations section helps; the problem is the abstract and conclusion still say 'high effectiveness,' which the evidence does not support.

This is a paper worth engaging seriously. A good referee would ask for a more careful evaluation: compare against existing summary methods, report variance across seeds, and add a no-bias control. The method deserves a shot; the claims need to shrink.

My recommendation: send to peer review, with a request for substantial revision on the evaluation.

Referee Report

4 major / 5 minor

Summary. The paper introduces FaceX, a post-hoc explanation method that aggregates per-image Grad-CAM attributions over 19 predefined facial regions to produce a single summary heatmap for a face attribute classifier, together with visualizations of the top-k image patches per region. The method is parameter-free apart from patch size and top-k choices, and is evaluated on CelebA and FairFace-trained models tested on CelebAMask-HQ and RFW, in controlled single- and multi-attribute bias injection scenarios, a bias mitigation comparison, and a real-data benchmark comparison. The central claim is that FaceX provides the first summary-level model explanation for face attribute classifiers and demonstrates high effectiveness at surfacing single- and multi-attribute biases.

Significance. If the effectiveness claim could be substantiated, FaceX would be a practically useful addition to the XAI toolbox for facial analysis, because it condenses many instance-level attribution maps into a compact, human-readable summary and is applicable to any model with access to region masks. The method is simple, transparent, and shipped with code, which are genuine strengths. However, the current evaluation does not yet establish that the IoR rankings measure model bias rather than generic activation patterns; the absence of baselines, control conditions, and variance estimates means the headline 'high effectiveness' claim rests on an unvalidated premise that the paper itself acknowledges in Section 4.2.

major comments (4)

[§4.2, Table 1] See Table 1 and Section 4.2.
[§5.2, Table 3] See Table 3 and Section 5.2.
[§5.3, Figure 4] See Section 5.3 and Figure 4.
[§5.4, Figure 5] See Section 5.4 and Figure 5.

minor comments (5)

[§3.3, Eq. (1)] Section 3.3, Eq. (1).
[§5.1] Section 5.1, text around Table 1.
[§2] Section 2, first paragraph of XAI subsection.
[§4.2] Section 4.2, implementation details.
[§5.5, Table 4] Section 5.5, Table 4.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: FaceX computes region-level attributions directly from model activations and is evaluated against a stated, falsifiable premise, not against its own outputs.

full rationale

FaceX's derivation chain is self-contained. The IoR values (Eq. 3-5) are a direct region-level average of Grad-CAM activations under face-parsing masks, and the high-impact patches (Eq. 6-7) are selected by thresholding the same activation values; no parameter is fitted to the evaluation targets, and no quantity called a 'prediction' is computed from the labels used to define bias. The evaluation assumes that a model biased toward an attribute will concentrate its activations in the corresponding facial region, so that the region's IoR ranks high. The paper explicitly states this premise 'might not align with the behavior of deep learning models' (Section 4.2), and Table 1 reports concrete failures (Eyeglasses at rank 6, Earrings at rank 5). This is a stated, empirical, falsifiable evaluation premise rather than a circular derivation: the method's output is not defined in terms of that premise, and the paper does not conceal the cases where the premise fails. The high-impact patches rely on the same IoR/Grad-CAM values and therefore cannot independently corroborate the ranking, but the paper does not claim such independent corroboration; this is a limitation of the evaluation design, not a circular inference. The self-citations are not load-bearing: FLAC [30] is used as an external bias-mitigation baseline, and the related-work citation [31] is contextual. No uniqueness theorem, ansatz, or forbidden-alternative argument is imported from the authors' prior work. Consequently, no specific circular step can be exhibited by quoting an equation or a fitted parameter that reduces to the target result.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

FaceX introduces no fitted parameters or new entities; the key unstated inputs are the patch size and Grad-CAM layer, and the main assumption is that region-level aggregation of Grad-CAM reveals bias. This assumption is clearly stated but not independently validated.

free parameters (2)

patch size Z
Used to partition images into patches for high-impact explanations (Section 3.5); the value is not reported in the paper, making exact replication harder.
number of top-k high-impact patches = 20
Table 2 reports the top 20 patches for each region, but the general k in Section 3.5 is not fixed; this is an experimental choice.

assumptions (4)

domain assumption FaRL provides accurate face parsing masks on arbitrary face images.
Section 3.2 uses FaRL to predict masks for datasets without ground-truth masks (e.g., RFW); inaccurate masks would distort region-level IoR values.
domain assumption Grad-CAM heatmaps faithfully reflect the model's decision-relevant regions.
Section 3.3 relies on Grad-CAM as the instance-level attribution method; if Grad-CAM is unfaithful, the aggregated summary would be misleading.
domain assumption The 19 facial regions from CelebAMask-HQ are sufficient to capture all potential bias-inducing attributes.
Section 3.2 defines the regions of interest based on CelebAMask-HQ; attributes not mapped to these regions (e.g., skin tone gradients) may be missed.
ad hoc to paper A model biased toward an attribute will concentrate its Grad-CAM activations in the corresponding facial region.
Section 4.2 makes this assumption to justify using ranking position as an evaluation metric; if false, FaceX's bias-detection signal would be unreliable.

how reviews work

0 comments

Cite this review

Pith. "Pith review of FaceX: Understanding Face Attribute Classifiers through Summary Model Explanations." pith.science (2026). https://pith.science/paper/GV7IX4LO

@misc{pith2026241207313,
  author       = {Pith},
  title        = {Pith review of: FaceX: Understanding Face Attribute Classifiers through Summary Model Explanations},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GV7IX4LO}},
  note         = {Machine review of arXiv:2412.07313}
}

read the original abstract

EXplainable Artificial Intelligence (XAI) approaches are widely applied for identifying fairness issues in Artificial Intelligence (AI) systems. However, in the context of facial analysis, existing XAI approaches, such as pixel attribution methods, offer explanations for individual images, posing challenges in assessing the overall behavior of a model, which would require labor-intensive manual inspection of a very large number of instances and leaving to the human the task of drawing a general impression of the model behavior from the individual outputs. Addressing this limitation, we introduce FaceX, the first method that provides a comprehensive understanding of face attribute classifiers through summary model explanations. Specifically, FaceX leverages the presence of distinct regions across all facial images to compute a region-level aggregation of model activations, allowing for the visualization of the model's region attribution across 19 predefined regions of interest in facial images, such as hair, ears, or skin. Beyond spatial explanations, FaceX enhances interpretability by visualizing specific image patches with the highest impact on the model's decisions for each facial region within a test benchmark. Through extensive evaluation in various experimental setups, including scenarios with or without intentional biases and mitigation efforts on four benchmarks, namely CelebA, FairFace, CelebAMask-HQ, and Racial Faces in the Wild, FaceX demonstrates high effectiveness in identifying the models' biases.

Figures

Figures reproduced from arXiv: 2412.07313 by the authors.

**Figure 1.** FaceX employs 19 facial regions and accessories to provide explanations (left: face regions, right: hat and glasses). Blue [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗

**Figure 2.** Grad-CAM instance-level explanations for six ran [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗

**Figure 3.** FaceX heatmap output for single attribute bias experiments on CelebA (train) and CelebAMask-HQ (test). Heatmap’s [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗

Figures from the paper (3 more)

**Figure 4.** Figure 4: FaceX heatmap output and high impact patches for [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]

**Figure 5.** Figure 5: Specifically, contrary to the initial training phase where [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]

**Figure 6.** Figure 6: FaceX heatmap outputs for experiments on de [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

45 extracted references · 30 canonical work pages

[1]

Reduan Achtibat, Maximilian Dreyer, Ilona Eisenbraun, Sebastian Bosse, Thomas Wiegand, Wojciech Samek, and Sebastian Lapuschkin. 2023. From attribution maps to human-understandable explanations through Concept Relevance Propa- gation. Nature Machine Intelligence 5, 9 (2023), 1006–1019

work page 2023
[2]

Amina Adadi and Mohammed Berrada. 2018. Peeking inside the black-box: a survey on explainable artificial intelligence (XAI). IEEE access 6 (2018), 52138– 52160

work page 2018
[3]

Joy Buolamwini and Timnit Gebru. 2018. Gender shades: Intersectional accu- racy disparities in commercial gender classification. In Conference on fairness, accountability and transparency. PMLR, 77–91

work page 2018
[4]

Aditya Chattopadhay, Anirban Sarkar, Prantik Howlader, and Vineeth N Balasub- ramanian. 2018. Grad-cam++: Generalized gradient-based visual explanations for deep convolutional networks. In 2018 IEEE winter conference on applications of computer vision (W ACV). IEEE, 839–847

work page 2018
[5]

Jessica Deuschel, Bettina Finzel, and Ines Rieger. 2020. Uncovering the bias in facial expressions. arXiv preprint arXiv:2011.11311 (2020)

arXiv 2020
[6]

Rachel Lea Draelos and Lawrence Carin. 2020. Use HiResCAM instead of Grad- CAM for faithful explanations of convolutional neural networks. arXiv preprint arXiv:2011.08891 (2020)

arXiv 2020
[7]

Simone Fabbrizzi, Symeon Papadopoulos, Eirini Ntoutsi, and Ioannis Kompat- siaris. 2022. A survey on bias in visual datasets. Computer Vision and Image Understanding 223 (2022), 103552

work page 2022
[8]

Anthony W Flores, Kristin Bechtel, and Christopher T Lowenkamp. 2016. False positives, false negatives, and false analyses: A rejoinder to machine bias: There’s software used across the country to predict future criminals. and it’s biased against blacks. Fed. Probation 80 (2016), 38

work page 2016

Show all 45 references

[9]

Amirata Ghorbani, James Wexler, James Y Zou, and Been Kim. 2019. Towards automatic concept-based explanations. Advances in neural information processing systems 32 (2019)

2019
[10]

Mark Ibrahim, Melissa Louie, Ceena Modarres, and John Paisley. 2019. Global explanations of neural networks: Mapping the landscape of predictions. In Pro- ceedings of the 2019 AAAI/ACM Conference on AI, Ethics, and Society . 279–287

2019
[11]

Peng-Tao Jiang, Chang-Bin Zhang, Qibin Hou, Ming-Ming Cheng, and Yunchao Wei. 2021. Layercam: Exploring hierarchical class activation maps for localization. IEEE Transactions on Image Processing 30 (2021), 5875–5888

2021
[12]

Jungseock Joo and Kimmo Kärkkäinen. 2020. Gender slopes: Counterfactual fairness for computer vision models by attribute manipulation. In Proceedings of 8 the 2nd international workshop on fairness, accountability, transparency and ethics in multimedia. 1–5

2020
[13]

Tushar Karayil, Philipp Blandfort, Jörn Hees, and Andreas Dengel. 2019. The focus-aspect-value model for explainable prediction of subjective visual inter- pretation. In Proceedings of the 2019 on International Conference on Multimedia Retrieval. 16–24

2019
[14]

Kimmo Karkkainen and Jungseock Joo. 2021. Fairface: Face attribute dataset for balanced race, gender, and age for bias measurement and mitigation. In Proceedings of the IEEE/CVF winter conference on applications of computer vision . 1548–1558

2021
[15]

Ashraf Khalil, Soha Glal Ahmed, Asad Masood Khattak, and Nabeel Al-Qirim
[16]

Anoop Krishnan, Ali Almadan, and Ajita Rattani. 2020. Understanding fairness of gender classification algorithms across gender-race groups. In 2020 19th IEEE International Conference on Machine Learning and Applications (ICMLA) . IEEE, 1028–1035

2020
[17]

Sebastian Lapuschkin, Stephan Wäldchen, Alexander Binder, Grégoire Montavon, Wojciech Samek, and Klaus-Robert Müller. 2019. Unmasking Clever Hans pre- dictors and assessing what machines really learn. Nature communications 10, 1 (2019), 1096

2019
[18]

Cheng-Han Lee, Ziwei Liu, Lingyun Wu, and Ping Luo. 2020. Maskgan: Towards diverse and interactive facial image manipulation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition . 5549–5558

2020
[19]

Brian Y Lim, Qian Yang, Ashraf M Abdul, and Danding Wang. 2019. Why these explanations? Selecting intelligibility types for explanation goals.. In IUI Workshops

2019
[20]

Mengchen Liu, Jiaxin Shi, Zhen Li, Chongxuan Li, Jun Zhu, and Shixia Liu. 2016. Towards better analysis of deep convolutional neural networks.IEEE transactions on visualization and computer graphics 23, 1 (2016), 91–100

2016
[21]

Ziwei Liu, Ping Luo, Xiaogang Wang, and Xiaoou Tang. 2015. Deep learning face attributes in the wild. In Proceedings of the IEEE international conference on computer vision. 3730–3738

2015
[22]

Scott M Lundberg and Su-In Lee. 2017. A unified approach to interpreting model predictions. Advances in neural information processing systems 30 (2017)

2017
[23]

Pietro Melzi, Ruben Tolosana, Ruben Vera-Rodriguez, Minchul Kim, Christian Rathgeb, Xiaoming Liu, Ivan DeAndres-Tame, Aythami Morales, Julian Fierrez, Javier Ortega-Garcia, et al. 2024. FRCSyn Challenge at WACV 2024: Face Recog- nition Challenge in the Era of Synthetic Data. I...

2024
[24]

Pietro Melzi, Ruben Tolosana, Ruben Vera-Rodriguez, Minchul Kim, Christian Rathgeb, Xiaoming Liu, Ivan DeAndres-Tame, Aythami Morales, Julian Fierrez, Javier Ortega-Garcia, et al. 2024. FRCSyn-onGoing: Benchmarking and compre- hensive evaluation of real and synthetic data to i...

2024
[25]

Grégoire Montavon, Alexander Binder, Sebastian Lapuschkin, Wojciech Samek, and Klaus-Robert Müller. 2019. Layer-wise relevance propagation: an overview. Explainable AI: interpreting, explaining and visualizing deep learning (2019), 193– 209

2019
[26]

Mei Ngan, Patrick J Grother, and Mei Ngan. 2015. Face recognition vendor test (FRVT) performance of automated gender classification algorithms . US Department of Commerce, National Institute of Standards and Technology

2015
[27]

Alejandro Pena, Ignacio Serna, Aythami Morales, and Julian Fierrez. 2020. Bias in multimodal AI: Testbed for fair automatic recruitment. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops . 28–29

2020
[28]

Why should i trust you?

Marco Tulio Ribeiro, Sameer Singh, and Carlos Guestrin. 2016. " Why should i trust you?" Explaining the predictions of any classifier. In Proceedings of the 22nd ACM SIGKDD international conference on knowledge discovery and data mining . 1135–1144

2016
[29]

Marco Tulio Ribeiro, Sameer Singh, and Carlos Guestrin. 2018. Anchors: High- precision model-agnostic explanations. In Proceedings of the AAAI conference on artificial intelligence, Vol. 32

2018
[30]

Ioannis Sarridis, Christos Koutlis, Symeon Papadopoulos, and Christos Diou. 2023. FLAC: Fairness-Aware Representation Learning by Suppressing Attribute-Class Associations. arXiv preprint arXiv:2304.14252 (2023)

2023 arXiv
[31]

Ioannis Sarridis, Christos Koutlis, Symeon Papadopoulos, and Christos Diou
[32]

Sven Schultze, Ani Withöft, Larbi Abdenebaoui, and Susanne Boll. 2023. Explain- ing Image Aesthetics Assessment: An Interactive Approach. In Proceedings of the 2023 ACM International Conference on Multimedia Retrieval . 20–28

2023
[33]

Ramprasaath R Selvaraju, Michael Cogswell, Abhishek Das, Ramakrishna Vedan- tam, Devi Parikh, and Dhruv Batra. 2017. Grad-cam: Visual explanations from deep networks via gradient-based localization. In Proceedings of the IEEE interna- tional conference on computer vision . 618–626

2017
[34]

Daniel Smilkov, Nikhil Thorat, Been Kim, Fernanda Viégas, and Martin Wat- tenberg. 2017. Smoothgrad: removing noise by adding noise. arXiv preprint arXiv:1706.03825 (2017)

2017 arXiv
[35]

Mukund Sundararajan, Ankur Taly, and Qiqi Yan. 2017. Axiomatic attribution for deep networks. In International conference on machine learning . PMLR, 3319– 3328

2017
[36]

Babak Taati, Shun Zhao, Ahmed B Ashraf, Azin Asgarian, M Erin Browne, Ken- neth M Prkachin, Alex Mihailidis, and Thomas Hadjistavropoulos. 2019. Algo- rithmic bias in clinical populations—evaluating and improving facial analysis technology in older adults with dementia. IEEE a...

2019
[37]

Yaniv Taigman, Ming Yang, Marc’Aurelio Ranzato, and Lior Wolf. 2014. Deepface: Closing the gap to human-level performance in face verification. In Proceedings of the IEEE conference on computer vision and pattern recognition . 1701–1708

2014
[38]

Haofan Wang, Zifan Wang, Mengnan Du, Fan Yang, Zijian Zhang, Sirui Ding, Piotr Mardziel, and Xia Hu. 2020. Score-CAM: Score-weighted visual explanations for convolutional neural networks. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition work...

2020
[39]

Mei Wang, Weihong Deng, Jiani Hu, Xunqiang Tao, and Yaohai Huang. 2019. Racial faces in the wild: Reducing racial bias by information maximization adap- tation network. In Proceedings of the ieee/cvf international conference on computer vision. 692–702

2019
[40]

Seyma Yucer, Furkan Tektas, Noura Al Moubayed, and Toby P Breckon. 2022. Measuring hidden bias within face recognition via racial phenotypes. In Pro- ceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision . 995–1004

2022
[41]

Quanshi Zhang, Wenguan Wang, and Song-Chun Zhu. 2018. Examining CNN representations with respect to dataset bias. InProceedings of the AAAI Conference on Artificial Intelligence, Vol. 32

2018
[42]

Sicheng Zhao, Yue Gao, Xiaolei Jiang, Hongxun Yao, Tat-Seng Chua, and Xi- aoshuai Sun. 2014. Exploring principles-of-art features for image emotion recog- nition. In Proceedings of the 22nd ACM international conference on Multimedia . 47–56

2014
[43]

Yinglin Zheng, Hao Yang, Ting Zhang, Jianmin Bao, Dongdong Chen, Yangyu Huang, Lu Yuan, Dong Chen, Ming Zeng, and Fang Wen. 2022. General facial rep- resentation learning in a visual-linguistic manner. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Re...

2022
[2020]

IEEE Access 8 (2020), 130751–130761

Investigating bias in facial analysis systems: A systematic review. IEEE Access 8 (2020), 130751–130761

2020
[2023]

arXiv preprint arXiv:2307.10011 (2023)

Towards Fair Face Verification: An In-depth Analysis of Demographic Biases. arXiv preprint arXiv:2307.10011 (2023)

2023 arXiv

Pith tools

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

[1] [1]

Reduan Achtibat, Maximilian Dreyer, Ilona Eisenbraun, Sebastian Bosse, Thomas Wiegand, Wojciech Samek, and Sebastian Lapuschkin. 2023. From attribution maps to human-understandable explanations through Concept Relevance Propa- gation. Nature Machine Intelligence 5, 9 (2023), 1006–1019

work page 2023

[2] [2]

Amina Adadi and Mohammed Berrada. 2018. Peeking inside the black-box: a survey on explainable artificial intelligence (XAI). IEEE access 6 (2018), 52138– 52160

work page 2018

[3] [3]

Joy Buolamwini and Timnit Gebru. 2018. Gender shades: Intersectional accu- racy disparities in commercial gender classification. In Conference on fairness, accountability and transparency. PMLR, 77–91

work page 2018

[4] [4]

Aditya Chattopadhay, Anirban Sarkar, Prantik Howlader, and Vineeth N Balasub- ramanian. 2018. Grad-cam++: Generalized gradient-based visual explanations for deep convolutional networks. In 2018 IEEE winter conference on applications of computer vision (W ACV). IEEE, 839–847

work page 2018

[5] [5]

Jessica Deuschel, Bettina Finzel, and Ines Rieger. 2020. Uncovering the bias in facial expressions. arXiv preprint arXiv:2011.11311 (2020)

arXiv 2020

[6] [6]

Rachel Lea Draelos and Lawrence Carin. 2020. Use HiResCAM instead of Grad- CAM for faithful explanations of convolutional neural networks. arXiv preprint arXiv:2011.08891 (2020)

arXiv 2020

[7] [7]

Simone Fabbrizzi, Symeon Papadopoulos, Eirini Ntoutsi, and Ioannis Kompat- siaris. 2022. A survey on bias in visual datasets. Computer Vision and Image Understanding 223 (2022), 103552

work page 2022

[8] [8]

Anthony W Flores, Kristin Bechtel, and Christopher T Lowenkamp. 2016. False positives, false negatives, and false analyses: A rejoinder to machine bias: There’s software used across the country to predict future criminals. and it’s biased against blacks. Fed. Probation 80 (2016), 38

work page 2016