Pith. sign in

REVIEW 4 major objections 6 minor 85 references

Seeing Through Deepfakes: A Human-Inspired Framework for Multi-Face Detection

T0 review · 4 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read This paper claims that multi-face deepfake detection improves when the detector is built around four contextual cues humans actually use: scene-motion coherence, inter-face appearance compatibility, interpersonal gaze alignment, and…

desk verdict A load-bearing fusion-rule contradiction means the headline gains are not tied to a well-defined model, but the human-study-driven framework is worth refereeing. read the letter →

arxiv 2507.14807 v1 pith:YE4CM55I submitted 2025-07-20 cs.CV cs.AI

Juan Hu , Shaojing Fan , Terence Sim This is my paper
classification cs.CVcs.AI
keywords deepfakedetectionmulti-facehumancognitioncontextualcuesframe-levelcompletegazealignmentface-bodyconsistencyinterpretability
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 sets out to show that detecting every fake face in a multi-face video is a different task from detecting one fake face, and that humans solve it by reading social context rather than inspecting faces in isolation. Through human studies, it identifies four cues people rely on: whether a face moves coherently with the scene, whether it matches the other faces in appearance, whether its gaze fits the group's gaze pattern, and whether its face agrees with its body in age and gender. The proposed system, HICOM, turns each cue into a module and fuses their outputs so that a frame is judged correct only if every face in it is classified correctly. On standard multi-face benchmarks the paper reports average frame-level accuracy gains of 3.3% in-dataset and 5.8% on an unseen dataset, plus human-readable explanations from a language model.

What carries the argument

The central object is HICOM, a four-module detector whose modules mirror the four human cues. M1 infers scene-motion coherence from multi-scale face and background features over time; M2 compares faces within a frame using a Transformer trained with contrastive and cross-entropy losses; M3 trains a gaze-locking classifier on eye regions and flags off-camera faces only when most faces in the scene face the camera; M4 trains age and gender classifiers separately on face crops and body crops and flags mismatches. A final XOR fusion marks a face as fake if any module flags it, so a single working cue can still catch a manipulated face.

What would settle it

Run a controlled study in which hundreds of naive viewers are shown newly generated multi-face deepfakes not drawn from the benchmark datasets and asked to state why faces look fake; if the four cues are not the dominant reasons they give, or if a version of HICOM with the gaze and body modules ablated matches the full model on those videos, the central claim is falsified.

Watch

Extended reading notes

Core claim

The central claim is that contextual inconsistency, not individual facial artifact, is the reliable signal for catching deepfake faces in social scenes. A two-phase human study yields four dominant cues, and HICOM encodes them as four modules: scene-motion coherence, inter-face appearance comparison, interpersonal gaze alignment, and face-body age and gender consistency. The paper reports that existing methods saturate on face-level accuracy but drop on frame-level complete multi-face detection, while HICOM improves average frame-level accuracy by 3.3% across FFIW, OpenForensics, and DF-Platter, maintains an edge under unseen perturbations, and generalizes to the unseen ManualFake dataset with a 5.8% average frame-level accuracy improvement.

Load-bearing premise

The claim depends on the assumption that the cues identified by a small group of human viewers on the benchmark datasets are the stable, transferable signals that drive detection, rather than quirks of those particular videos or viewers.

Editorial extensions

If this is right

  • Frame-level complete multi-face detection becomes a metric that separates methods, since face-level accuracy alone is nearly saturated across baselines.
  • Human-inspired contextual cues transfer to unseen datasets and perturbations better than artifact-specific features, as shown by the 5.8% gain on the untrained ManualFake dataset.
  • The modular design means a newly discovered human cue can be added as an extra module without retraining the entire system.
  • XOR fusion makes detection robust in the sense that one reliable cue is sufficient to flag a fake face even when the other modules fail.
  • An attached language model can turn module scores into plain-language explanations for each face, making the system's verdicts inspectable by non-experts.

Reading between the lines

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

  • Editorial inference: if the four cues are genuinely stable across deepfake generations, detectors built on them should degrade more gracefully on future generation methods than detectors trained on pixel-level artifacts.
  • Editorial inference: the gaze module's decision rule presupposes that most faces in a scene look at the camera; in candid or surveillance footage where nobody does, this cue goes silent, so the framework would need a different social baseline for such scenes.
  • Editorial inference: the paper's cue-prevalence numbers could be tested against learned fusion weights, which would reveal whether the human-reported proportions (motion 34.2%, appearance 31.5%, gaze 25.0%, body 7.5%) are also the optimal ones for the model.
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

4 major / 6 minor

Summary. The paper proposes HICOM, a multi-face deepfake detection framework whose design is motivated by a two-phase human study. The authors report four human cues (scene-motion coherence, inter-face appearance compatibility, interpersonal gaze alignment, and face-body consistency), instantiate each as a module (M1–M4), and fuse module outputs to label every face in a frame. Experiments on FFIW, OpenForensics, DF-Platter, and ManualFake report improvements in frame-level complete multi-face detection accuracy, robustness to perturbations, cross-dataset generalization, and an LLM-based explanation interface.

Significance. The paper's central idea—basing multi-face detection on cues elicited from human observers and evaluating a strict frame-level complete multi-face metric—is timely and potentially useful. Strengths include a clearly defined evaluation metric, a modular architecture with each module tied to a stated hypothesis, and comparisons against multiple single- and multi-face baselines. If the cue discovery and fusion specification are made rigorous and non-circular, the approach could make a solid contribution to the deepfake detection literature. At present, however, the reported gains rest on an internally contradictory fusion rule and on cue-selection evidence that is confounded with the evaluation benchmarks.

major comments (4)
  1. [§4.6] The fusion rule as written is self-contradictory. The text says the module outputs are combined 'using an XOR operation, ensuring that any detected anomaly leads to a fake face prediction,' but XOR outputs 1 only when an odd number of inputs are 1; a face flagged by exactly two modules yields 1 XOR 1 = 0, i.e., real. The stated rationale ('When M1 and M2 miss a fake face, M3 and M4 help identify these inconsistencies') and the monotone ablation in Table 4 (M1 < M1+M2 < M1+M2+M3 < M1+M2+M3+M4) both describe an OR-like rule rather than XOR. Since the fused output is the final prediction behind Tables 1–3, the reported 3.3% in-dataset and 5.8% cross-dataset gains cannot be attributed to the proposed cues unless the actual fusion rule is specified (including how each module's continuous output is binarized, any thresholds, and the exact Boolean combination) and shown to reproduce the tables. This is a load-bearing specification, not a wording issue.
  2. [§3.2, §5.3] The cue discovery is circular with the evaluation. The four cues were elicited on the same OpenForensics, FFIW, and DF-Platter datasets used to evaluate HICOM in Tables 1 and 2; the cross-dataset test in Table 3 uses ManualFake, which was not part of cue discovery, but the design choices (which cues to include, how to threshold them) were already informed by the evaluation distribution. As a result, the 5.8% generalization gain cannot be cleanly attributed to human-inspired cues as opposed to cues that happen to be informative on these benchmarks. Please validate the cues on a held-out split or an independent deepfake type not seen during cue discovery, and at minimum explicitly discuss this confound in the limitations.
  3. [§3.2, §3.3] The quantitative evidence for the four cues is thin. Phase 1 recruited only four participants, each reviewing 500 samples, and §3.3 reports prevalence percentages (34.2%, 31.5%, 25.0%, 7.5%) without per-participant variance, inter-rater reliability, or significance tests. The criterion 'Only identifications matching dataset labels were considered valid' further conditions the cue reporting on label consistency, which may suppress cues that lead to false alarms. With n=4, these percentages cannot support the claim that the four cues are the key human strategies. Please report the full distribution of cue frequencies across participants, agreement metrics, and the number of samples contributing to each percentage.
  4. [§4.4, Eq. (3)] The gaze module's decision rule is ad hoc and not grounded in the human study. The conditions 'n_L - n_O > 1 or n_T = 2' appear without derivation or ablation; similarly, the contrastive margin and λcomp in Eq. (2) are set empirically. Since Table 4 attributes incremental gains to M3 and M4, these hand-set thresholds could be the source of the improvement rather than the human-inspired cue itself. Please ablate these thresholds and report sensitivity, or justify them from the human-study data.
minor comments (6)
  1. [Table 1] The table header contains the typo 'OpenForencics'; it should read 'OpenForensics'.
  2. [§4.6] The text refers to a Supplementary Material for the fusion strategy, but the submitted manuscript does not appear to include that supplement; either provide it or remove the reference.
  3. [§5.4, Fig. 5] The human-comparison result is reported only as 'three stars'; please give the actual test statistic, p-value, and sample size.
  4. [Related Work] Reference [23] is cited for the 'Dual-path Actor Interaction framework with Multi-scale Actor Contrastive Loss,' which does not match the Gallagher and Chen paper; the citation appears to be incorrect.
  5. [Abstract / §5.2] The abstract says 'average accuracy by 3.3%,' while the body specifies 'average improvements of 3.3% in FCAC, and 3.1% in FCAU'; clarify which metric is being summarized.
  6. [§3.2] Phase 2 reports 20 MTurk participants but does not state the number of judgments per item, compensation, or exclusion criteria; adding these details would improve reproducibility.

Circularity Check

1 steps flagged · score 4.0 of 10

Human-study cues are extracted from the same three benchmarks used for in-dataset evaluation, so the headline 3.3% gain is partly circular; the ManualFake test preserves independent content.

  1. fitted input called prediction [Section 3.2 (human study) and Section 5.2 (in-dataset evaluation); cf. Limitations in Section 6]
    "In the first phase, we randomly selected 2, 000 multi-face deepfake videos and images from the OpenForensics [39], FFIW [84], and DF-Platter [53]... Only identifications matching the dataset labels were considered valid."

    The four cues that define HICOM's four modules were elicited from the same three datasets (OpenForensics, FFIW, DF-Platter) used for the in-dataset evaluation in Table 1, and participant reports were filtered by those datasets' labels. Therefore the claim that HICOM improves average accuracy by 3.3% in Section 5.2 measures performance on the distribution from which the cue set was selected; it is a discovery-set evaluation rather than a test of cue transfer. The conclusion's limitation, 'Our findings are based on all benchmark multi-face deepfake datasets,' concedes the benchmark-bound origin of the cues.

full rationale

Most of HICOM's engineering is self-contained: the modules are standard networks trained with cross-entropy and contrastive losses, and the comparisons to SOTA baselines are external. The human-study cue selection, however, is not distribution-independent. Phase 1 sampled 2,000 items from OpenForensics, FFIW, and DF-Platter and kept only identifications agreeing with those datasets' labels; the resulting four cues then define the four modules. Section 5.2 reports the headline 3.3% in-dataset FCAC improvement on exactly those three datasets. So the in-dataset result evaluates the model on the distribution from which its architectural cues were elicited; the paper's own limitation notes that the findings are based on all benchmark multi-face deepfake datasets. This is a partial, not total, circularity: the cue-to-module mapping still had to be implemented and trained, and the ManualFake cross-dataset experiment in Section 5.3 explicitly excludes ManualFake from the human study and training, giving independent evidence for transfer. The XOR fusion description in Section 4.6 is internally inconsistent with 'ensuring that any detected anomaly leads to a fake face prediction,' but that is a correctness/consistency defect rather than a circularity. Author self-citations ([18], [32]) appear in related-work and motivation contexts and are not load-bearing. Overall score 4.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The central claim relies on a set of hand-set hyperparameters, an unspecified fusion rule, and several domain assumptions about human perception and dataset labels. There are no newly invented physical or conceptual entities beyond the design choices of the framework.

free parameters (5)
  • contrastive margin = 1.0
    Set as default in Eq. 2 without sensitivity analysis.
  • lambda_comp = 0.3
    Empirically set in Eq. 2 to weight the contrastive loss.
  • gaze decision thresholds = nL - nO > 1 or nT = 2
    Hand-crafted rule in Eq. 3 for flagging abnormal gaze, with no justification for these exact conditions.
  • module fusion rule = XOR stated, OR described
    Section 4.6 says XOR but describes 'any detected anomaly leads to fake', which is OR. This inconsistency is not resolved.
  • age/gender categories = child, middle-aged, senior; male, female
    Chosen as output classes for M4 without evidence that these categories are optimal or sufficient.
assumptions (4)
  • domain assumption Human cognitive patterns in face perception are stable and transferable to AI detection.
    Stated in Section 1 as the rationale for building on human studies; not independently validated beyond the datasets used.
  • domain assumption Dataset labels in OpenForensics, FFIW, DF-Platter, and ManualFake are accurate ground truth for deepfake presence.
    Used to validate human identifications in Section 3.2 and to train and evaluate all models; label errors would propagate into all results.
  • ad hoc to paper The four identified cues are the dominant cues and minor cues can be excluded.
    Section 3.3 excludes minor cues (1.8%) and integrates only four cues, without validation that the excluded cues are irrelevant.
  • domain assumption IMDB-WIKI provides a suitable basis for age/gender classification of faces and bodies in these datasets.
    Used in Section 4.5 to train age and gender models; domain shift between IMDB-WIKI and the evaluation datasets is not analyzed.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Seeing Through Deepfakes: A Human-Inspired Framework for Multi-Face Detection." pith.science (2026). https://pith.science/paper/YE4CM55I

@misc{pith2026250714807,
  author       = {Pith},
  title        = {Pith review of: Seeing Through Deepfakes: A Human-Inspired Framework for Multi-Face Detection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YE4CM55I}},
  note         = {Machine review of arXiv:2507.14807}
}
read the original abstract

Multi-face deepfake videos are becoming increasingly prevalent, often appearing in natural social settings that challenge existing detection methods. Most current approaches excel at single-face detection but struggle in multi-face scenarios, due to a lack of awareness of crucial contextual cues. In this work, we develop a novel approach that leverages human cognition to analyze and defend against multi-face deepfake videos. Through a series of human studies, we systematically examine how people detect deepfake faces in social settings. Our quantitative analysis reveals four key cues humans rely on: scene-motion coherence, inter-face appearance compatibility, interpersonal gaze alignment, and face-body consistency. Guided by these insights, we introduce \textsf{HICOM}, a novel framework designed to detect every fake face in multi-face scenarios. Extensive experiments on benchmark datasets show that \textsf{HICOM} improves average accuracy by 3.3\% in in-dataset detection and 2.8\% under real-world perturbations. Moreover, it outperforms existing methods by 5.8\% on unseen datasets, demonstrating the generalization of human-inspired cues. \textsf{HICOM} further enhances interpretability by incorporating an LLM to provide human-readable explanations, making detection results more transparent and convincing. Our work sheds light on involving human factors to enhance defense against deepfakes.

Figures

Figures reproduced from arXiv: 2507.14807 by the authors.

Figure 1
Figure 1. This work takes a novel human-centric approach to [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Examples of human studies with and without the four [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. HICOM leverages human-inspired cues (H1−H4) derived from human studies to detect all fake faces within multi-face settings. integrates human-derived reasons. Unlike methods that merely superimpose models, each module of HICOM is in￾spired by specific human-reported cues. According to the H1 (scene-motion coherence), we develop a module that in￾tegrates facial and contextual features from preceding and following fram… view at source ↗
Figures from the paper (2 more)
Figure 5
Figure 5. Figure 5: HICOM surpasses humans in multi-face detection. HICOM, which shows our model’s potential to work with LLM to provide an explainable prediction. Results in [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 4
Figure 4. Figure 4: HICOM provides comprehensible explanations for its predictions through integration with an LLM. face scenarios. Specifically, we modify M1 to extract only scene￾motion features and M2 to focus solely on single-face features, re￾moving inter-face dependencies. M3 is exc…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

85 extracted references · 74 canonical work pages

  1. [1]

    Gpt-4 technical report

    Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ah- mad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, et al. Gpt-4 technical report. Open AI, 2023. 8

    2023

  2. [2]

    Mesonet: a compact facial video forgery detection network

    Darius Afchar, Vincent Nozick, Junichi Yamagishi, and Isao Echizen. Mesonet: a compact facial video forgery detection network. In WIFS, pages 1–7, 2018. 3

    2018

  3. [3]

    Limits of deepfake detection: A robust estimation viewpoint

    Sakshi Agarwal and Lav R Varshney. Limits of deepfake detection: A robust estimation viewpoint. arXiv preprint arXiv:1905.03493, 2019. 3

    arXiv 1905

  4. [4]

    Image forgery detection by trans- forming local descriptors into deep-derived features.Applied Soft Computing, 147:110730, 2023

    Muhammad Aqib Anwar, Syed Fahad Tahir, Labiba Gillani Fahad, and Kashif Kifayat. Image forgery detection by trans- forming local descriptors into deep-derived features.Applied Soft Computing, 147:110730, 2023. 6

    2023

  5. [5]

    Exposing the deception: Uncov- ering more forgery clues for deepfake detection

    Zhongjie Ba, Qingyu Liu, Zhenguang Liu, Shuang Wu, Feng Lin, Li Lu, and Kui Ren. Exposing the deception: Uncov- ering more forgery clues for deepfake detection. In AAAI, pages 719–728, 2024. 2, 3

    2024

  6. [6]

    Zelensky told to leave white house after angry spat with trump and vance

    BBC. Zelensky told to leave white house after angry spat with trump and vance. https://www.bbc.com/news/ live/c625ex282zzt, 2025. Accessed: 2025-03-07. 1

    2025

  7. [7]

    End-to-end reconstruction- classification learning for face forgery detection

    Junyi Cao, Chao Ma, Taiping Yao, Shen Chen, Shouhong Ding, and Xiaokang Yang. End-to-end reconstruction- classification learning for face forgery detection. In CVPR, pages 4113–4122, 2022. 3

    work page 2022

  8. [8]

    Unveiling the truth: Exploring human gaze patterns in fake images

    Giuseppe Cartella, Vittorio Cuculo, Marcella Cornia, and Rita Cucchiara. Unveiling the truth: Exploring human gaze patterns in fake images. IEEE Signal Processing Letters ,

    work page

Show all 85 references
  1. [9]

    Role of human physiology and facial biomechanics towards building robust deepfake detectors: A comprehensive survey and analysis

    Rajat Chakraborty and Ruchira Naskar. Role of human physiology and facial biomechanics towards building robust deepfake detectors: A comprehensive survey and analysis. Computer Science Review, 54:100677, 2024. 4

    2024

  2. [10]

    Spar- tan: Self-supervised spatiotemporal transformers approach to group activity recognition

    Naga VS Chappa, Pha Nguyen, Alexander H Nelson, Han- Seok Seo, Xin Li, Page Daniel Dobbs, and Khoa Luu. Spar- tan: Self-supervised spatiotemporal transformers approach to group activity recognition. In CVPR, pages 5158–5168,

  3. [11]

    Simswap: An efficient framework for high fidelity face swapping

    Renwang Chen, Xuanhong Chen, Bingbing Ni, and Yanhao Ge. Simswap: An efficient framework for high fidelity face swapping. In ACM MM, pages 2003–2011, 2020. 3

    2003

  4. [12]

    Unsupervised outlier detection in appearance-based gaze es- timation

    Zhaokang Chen, Didan Deng, Jimin Pi, and Bertram E Shi. Unsupervised outlier detection in appearance-based gaze es- timation. In ICCVW, 2019. 6

    2019

  5. [13]

    Exploiting style latent flows for generalizing deepfake video detection

    Jongwook Choi, Taehoon Kim, Yonghyun Jeong, Seungryul Baek, and Jongwon Choi. Exploiting style latent flows for generalizing deepfake video detection. In CVPR, pages 1133–1143, 2024. 2, 3

    2024

  6. [14]

    Fakecatcher: Detection of synthetic portrait videos using biological sig- nals

    Umur Aybars Ciftci, Ilke Demir, and Lijun Yin. Fakecatcher: Detection of synthetic portrait videos using biological sig- nals. TPAMI, 2020. 3

    2020

  7. [15]

    Evaluating amazon’s mechanical turk as a tool for experimental behavioral research

    Matthew JC Crump, John V McDonnell, and Todd M Gureckis. Evaluating amazon’s mechanical turk as a tool for experimental behavioral research. PloS one , 8(3):e57410,

  8. [16]

    Body image: Gender, ethnic, and age differences

    Jack Demarest and Rita Allen. Body image: Gender, ethnic, and age differences. The Journal of Social Psychology, 140 (4):465–472, 2000. 4

    2000

  9. [17]

    The deepfake detection challenge (dfdc) dataset

    Brian Dolhansky, Joanna Bitton, Ben Pflaum, Jikuo Lu, Russ Howes, Menglin Wang, and Cristian Canton Ferrer. The deepfake detection challenge (dfdc) dataset. arXiv preprint arXiv:2006.07397, 2020. 4

    2006 arXiv

  10. [18]

    Human per- ception of visual realism for photo and computer-generated face images

    Shaojing Fan, Rangding Wang, Tian-Tsong Ng, Cheston Y-C Tan, Jonathan S Herberg, and Bryan L Koenig. Human per- ception of visual realism for photo and computer-generated face images. ACM TAP, 11(2):1–21, 2014. 3

    2014

  11. [19]

    What is” special” about face perception? Psychological review, 105(3):482, 1998

    Martha J Farah, Kevin D Wilson, Maxwell Drain, and James N Tanaka. What is” special” about face perception? Psychological review, 105(3):482, 1998. 2

    1998

  12. [20]

    Creating, using, misusing, and detecting deep fakes

    Hany Farid. Creating, using, misusing, and detecting deep fakes. Journal of Online Trust and Safety, 1(4), 2022. 3

    2022

  13. [21]

    Recognition of images de- graded by gaussian blur

    Jan Flusser, Sajad Farokhi, Cyril H ¨oschl, Tom´aˇs Suk, Bar- bara Zitova, and Matteo Pedone. Recognition of images de- graded by gaussian blur. TIP, 25(2):790–806, 2015. 6

    2015

  14. [22]

    Social belonging motivates categoriza- tion of racially ambiguous faces

    Sarah E Gaither, Kristin Pauker, Michael L Slepian, and Samuel R Sommers. Social belonging motivates categoriza- tion of racially ambiguous faces. Social cognition, 34(2): 97–118, 2016. 2

    2016

  15. [23]

    Understanding im- ages of groups of people

    Andrew C Gallagher and Tsuhan Chen. Understanding im- ages of groups of people. In CVPR, pages 256–263, 2009. 2, 3

    2009

  16. [24]

    Automatic gaze analysis: A survey of deep learning based approaches

    Shreya Ghosh, Abhinav Dhall, Munawar Hayat, Jarrod Knibbe, and Qiang Ji. Automatic gaze analysis: A survey of deep learning based approaches. TPAMI, 46(1):61–84, 2023. 6

    2023

  17. [25]

    Delving into the local: Dynamic in- consistency learning for deepfake video detection

    Zhihao Gu, Yang Chen, Taiping Yao, Shouhong Ding, Jilin Li, and Lizhuang Ma. Delving into the local: Dynamic in- consistency learning for deepfake video detection. In AAAI, pages 744–752, 2022. 3

    2022

  18. [26]

    Ex- ploring spatial-temporal features for deepfake detection and localization

    Wu Haiwei, Zhou Jiantao, Zhang Shile, and Tian Jinyu. Ex- ploring spatial-temporal features for deepfake detection and localization. arXiv preprint arXiv:2210.15872, 2022. 3, 6

    2022 arXiv

  19. [27]

    Lips don’t lie: A generalisable and robust approach to face forgery detection

    Alexandros Haliassos, Konstantinos V ougioukas, Stavros Petridis, and Maja Pantic. Lips don’t lie: A generalisable and robust approach to face forgery detection. In CVPR, pages 5039–5049, 2021. 3

    2021

  20. [28]

    Haxby, Elizabeth A

    James V . Haxby, Elizabeth A. Hoffman, and Maria I. Gob- bini. The distributed human neural system for face percep- tion. Trends in Cognitive Sciences, 4(6):223–233, 2000. 3

    2000

  21. [29]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In CVPR, pages 770–778, 2016. 5

    2016

  22. [30]

    Mask r-cnn

    Kaiming He, Georgia Gkioxari, Piotr Doll ´ar, and Ross Gir- shick. Mask r-cnn. In ICCV, pages 2961–2969, 2017. 5

    2017

  23. [31]

    How does gaze to faces support face-to-face interaction? a review and perspective

    Roy S Hessels. How does gaze to faces support face-to-face interaction? a review and perspective. Psychonomic Bulletin & Review, 27(5):856–881, 2020. 4 9

    2020

  24. [32]

    Detecting compressed deepfake videos in social networks using frame- temporality two-stream convolutional network

    Juan Hu, Xin Liao, Wei Wang, and Zheng Qin. Detecting compressed deepfake videos in social networks using frame- temporality two-stream convolutional network. TCSVT, 32 (3):1089–1102, 2021. 3

    2021

  25. [33]

    A hierarchical deep temporal model for group activity recognition

    Mostafa S Ibrahim, Srikanth Muralidharan, Zhiwei Deng, Arash Vahdat, and Greg Mori. A hierarchical deep temporal model for group activity recognition. In CVPR, pages 1971– 1980, 2016. 4

    1971

  26. [34]

    Double face: Leveraging user intelligence to characterize and recognize ai-synthesized faces

    Matthew Joslin, Xian Wang, and Shuang Hao. Double face: Leveraging user intelligence to characterize and recognize ai-synthesized faces. In USENIX Security, pages 1009–1026,

  27. [35]

    Domain specificity in face perception

    Nancy Kanwisher. Domain specificity in face perception. Nature neuroscience, 3(8):759–763, 2000. 2

    2000

  28. [36]

    Nancy Kanwisher, Josh McDermott, and Marvin M. Chun. The fusiform face area: A module in human extrastriate cor- tex specialized for face perception. The Journal of Neuro- science, 17(11):4302–4311, 1997. 3

    1997

  29. [37]

    Can you all look here? towards determining gaze uniformity in group images

    Omkar N Kulkarni, Vikram Patil, Shivam B Parikh, Shashank Arora, and Pradeep K Atrey. Can you all look here? towards determining gaze uniformity in group images. In ISM, pages 100–103, 2020. 6

    2020

  30. [38]

    Do the eyes have it? cues to the direction of social attention

    Stephen RH Langton, Roger J Watt, and Vicki Bruce. Do the eyes have it? cues to the direction of social attention. Trends in Cognitive Sciences, 4(2):50–59, 2000. 4

    2000

  31. [39]

    Openforensics: Large-scale challenging dataset for multi-face forgery detection and segmentation in- the-wild

    Trung-Nghia Le, Huy H Nguyen, Junichi Yamagishi, and Isao Echizen. Openforensics: Large-scale challenging dataset for multi-face forgery detection and segmentation in- the-wild. In ICCV, pages 10117–10127, 2021. 3, 4, 6

    2021

  32. [40]

    Faceshifter: Towards high fidelity and occlusion aware face swapping

    Lingzhi Li, Jianmin Bao, Hao Yang, Dong Chen, and Fang Wen. Faceshifter: Towards high fidelity and occlusion aware face swapping. arXiv preprint arXiv:1912.13457, 2019. 3

    1912 arXiv

  33. [41]

    Sharp mul- tiple instance learning for deepfake video detection

    Xiaodan Li, Yining Lang, Yuefeng Chen, Xiaofeng Mao, Yuan He, Shuhui Wang, Hui Xue, and Quan Lu. Sharp mul- tiple instance learning for deepfake video detection. In ACM MM, pages 1864–1872, 2020. 2, 3

    2020

  34. [42]

    Exposing Deepfake videos by detecting face warping artifacts

    Yuezun Li and Siwei Lyu. Exposing Deepfake videos by detecting face warping artifacts. In CVPRW, pages 46–52,

  35. [43]

    Exploiting facial relationships and feature aggregation for multi-face forgery detection.TIFS, 19:8832– 8844, 2024

    Chenhao Lin, Fangbin Yi, Hang Wang, Qian Li, Deng Jingyi, and Chao Shen. Exploiting facial relationships and feature aggregation for multi-face forgery detection.TIFS, 19:8832– 8844, 2024. 2, 3, 7, 8

    2024

  36. [44]

    Preserving fairness generalization in deepfake detection

    Li Lin, Xinan He, Yan Ju, Xin Wang, Feng Ding, and Shu Hu. Preserving fairness generalization in deepfake detection. In CVPR, pages 16815–16825, 2024. 2, 7, 8

    2024

  37. [45]

    Spatial- phase shallow learning: rethinking face forgery detection in frequency domain

    Honggu Liu, Xiaodan Li, Wenbo Zhou, Yuefeng Chen, Yuan He, Hui Xue, Weiming Zhang, and Nenghai Yu. Spatial- phase shallow learning: rethinking face forgery detection in frequency domain. In CVPR, pages 772–781, 2021. 3

    2021

  38. [46]

    Exposingaicreated fakevideosbydetectingeyeblinking

    Chang M Liy and LYUS InIctuOculi. Exposingaicreated fakevideosbydetectingeyeblinking. In WIFS, 2018. 4

    2018

  39. [47]

    Accurate and time-saving deepfake detection in multi-face scenarios using combined features

    Zekun Ma and Bin Liu. Accurate and time-saving deepfake detection in multi-face scenarios using combined features. In ICCSSE, pages 378–382, 2022. 2, 3, 7, 8

    2022

  40. [48]

    Two- branch recurrent network for isolating deepfakes in videos

    Iacopo Masi, Aditya Killekar, Royston Marian Mascaren- has, Shenoy Pratik Gurudatt, and Wael AbdAlmageed. Two- branch recurrent network for isolating deepfakes in videos. In ECCV, pages 667–684, 2020. 3

    2020

  41. [49]

    Ex- ploiting visual artifacts to expose deepfakes and face manip- ulations

    Florian Matern, Christian Riess, and Marc Stamminger. Ex- ploiting visual artifacts to expose deepfakes and face manip- ulations. CVPRW, pages 288–295, 2019. 3

    2019

  42. [50]

    Mixture-of-noises enhanced forgery-aware predictor for multi-face manipulation detection and localization

    Changtao Miao, Qi Chu, Tao Gong, Zhentao Tan, Zhenchao Jin, Wanyi Zhuang, Man Luo, Honggang Hu, and Nenghai Yu. Mixture-of-noises enhanced forgery-aware predictor for multi-face manipulation detection and localization. arXiv preprint arXiv:2408.02306, 2024. 2, 3, 7, 8

    2024 arXiv

  43. [51]

    I hardly lie: A multistage fake news de- tection system

    Suchintan Mishra, Harshit Raj Sinha, Tushar Mitra, and Manadeepa Sahoo. I hardly lie: A multistage fake news de- tection system. In Biologically Inspired Techniques in Many Criteria Decision Making: Proceedings of BITMDM 2021 , pages 253–261. Springer, 2022. 6

    2021

  44. [52]

    Emotions don’t lie: An audio- visual deepfake detection method using affective cues

    Trisha Mittal, Uttaran Bhattacharya, Rohan Chandra, Aniket Bera, and Dinesh Manocha. Emotions don’t lie: An audio- visual deepfake detection method using affective cues. In ACM MM, pages 2823–2832, 2020. 3

    2020

  45. [53]

    Df-platter: Multi- face heterogeneous deepfake dataset

    Kartik Narayan, Harsh Agarwal, Kartik Thakral, Surbhi Mit- tal, Mayank Vatsa, and Richa Singh. Df-platter: Multi- face heterogeneous deepfake dataset. In CVPR, pages 9739– 9748, 2023. 3, 4, 6

    2023

  46. [54]

    Nightingale and Hany Farid

    Sophia J. Nightingale and Hany Farid. Ai-synthesized faces are indistinguishable from real faces and more trustworthy. Proceedings of the National Academy of Sciences , 119(8): e2120481119, 2022. 3

    2022

  47. [55]

    Fsgan: Subject agnostic face swapping and reenactment

    Yuval Nirkin, Yosi Keller, and Tal Hassner. Fsgan: Subject agnostic face swapping and reenactment. In ICCV, pages 7184–7193, 2019. 3

    2019

  48. [56]

    Deepfakes in video group

    PBS NEWS. Deepfakes in video group. https://www. channelnewsasia.com/commentary/deepfake- scam - video - conference - zoom - hong - kong - employee-4103266, 2023. Accessed: 2024-08-05. 1

    2023

  49. [57]

    Pudd: Towards robust multi-modal prototype-based deepfake de- tection

    Alvaro Lopez Pellicer, Yi Li, and Plamen Angelov. Pudd: Towards robust multi-modal prototype-based deepfake de- tection. In CVPR, pages 3809–3817, 2024. 2, 3

    2024

  50. [58]

    Deepfacelab: In- tegrated, flexible and extensible face-swapping framework

    Ivan Perov, Daiheng Gao, Nikolay Chervoniy, Kunlin Liu, Sugasa Marangonda, Chris Um ´e, Mr Dpfks, Carl Shift Facenheim, Luis RP, Jian Jiang, et al. Deepfacelab: In- tegrated, flexible and extensible face-swapping framework. arXiv preprint arXiv:2005.05535, 2020. 3

    2005 arXiv

  51. [59]

    Adversarial latent autoencoders

    Stanislav Pidhorskyi, Donald A Adjeroh, and Gianfranco Doretto. Adversarial latent autoencoders. In CVPR, pages 14104–14113, 2020. 3

    2020

  52. [60]

    Thinking in frequency: Face forgery detection by min- ing frequency-aware clues

    Yuyang Qian, Guojun Yin, Lu Sheng, Zixuan Chen, and Jing Shao. Thinking in frequency: Face forgery detection by min- ing frequency-aware clues. In ECCV, pages 86–103, 2020. 3

    2020

  53. [61]

    Age and gender recog- nition in the wild with deep attention

    Pau Rodr ´ıguez, Guillem Cucurull, Josep M Gonfaus, F Xavier Roca, and Jordi Gonzalez. Age and gender recog- nition in the wild with deep attention. Pattern Recognition, 72:563–571, 2017. 3

    2017

  54. [62]

    Faceforen- sics++: Learning to detect manipulated facial images

    Andreas Rossler, Davide Cozzolino, Luisa Verdoliva, Chris- tian Riess, Justus Thies, and Matthias Nießner. Faceforen- sics++: Learning to detect manipulated facial images. In ICCV, pages 1–11, 2019. 8 10

    2019

  55. [63]

    Dex: Deep expectation of apparent age from a single image

    Rasmus Rothe, Radu Timofte, and Luc Van Gool. Dex: Deep expectation of apparent age from a single image. In ICCVW, pages 10–15, 2015. 6

    2015

  56. [64]

    A review of driver gaze estimation and application in gaze behavior understanding

    Pavan Kumar Sharma and Pranamesh Chakraborty. A review of driver gaze estimation and application in gaze behavior understanding. Engineering Applications of Artificial Intel- ligence, 133:108117, 2024. 4

    2024

  57. [65]

    Scale-aware cnn for crowd density estimation and crowd behavior analysis

    Vipal Kumar Sharma, Roohie Naaz Mir, and Chandrapal Singh. Scale-aware cnn for crowd density estimation and crowd behavior analysis. Computers and Electrical Engi- neering, 106:108569, 2023. 4

    2023

  58. [66]

    Spatio-temporal graph representation learning for fraudster group detection

    Saeedreza Shehnepoor, Roberto Togneri, Wei Liu, and Mo- hammed Bennamoun. Spatio-temporal graph representation learning for fraudster group detection. TNNLS, 2022. 4

    2022

  59. [67]

    Inter- preting the latent space of gans for semantic face editing

    Yujun Shen, Jinjin Gu, Xiaoou Tang, and Bolei Zhou. Inter- preting the latent space of gans for semantic face editing. In CVPR, pages 9243–9252, 2020. 3

    2020

  60. [68]

    Detecting deep- fakes with self-blended images

    Kaede Shiohara and Toshihiko Yamasaki. Detecting deep- fakes with self-blended images. In CVPR, pages 18720– 18729, 2022. 7, 8

    2022

  61. [69]

    Gaze locking: passive eye contact detection for human- object interaction

    Brian A Smith, Qi Yin, Steven K Feiner, and Shree K Na- yar. Gaze locking: passive eye contact detection for human- object interaction. In ACM Symposium on UIST, pages 271– 280, 2013. 6

    2013

  62. [70]

    Deepfake video detection via facial action dependencies estimation

    Lingfeng Tan, Yunhong Wang, Junfu Wang, Liang Yang, Xunxun Chen, and Yuanfang Guo. Deepfake video detection via facial action dependencies estimation. In AAAI, pages 5276–5284, 2023. 2, 3

    2023

  63. [71]

    Illumination enlightened spatial-temporal inconsistency for deepfake video detection

    Kaiyue Tian, Chen Chen, Yichao Zhou, and Xiyuan Hu. Illumination enlightened spatial-temporal inconsistency for deepfake video detection. In ICME, pages 1–6. IEEE, 2024. 3

    2024

  64. [72]

    Fake ai videos about trump fights with zelenskyy

    Tiktok. Fake ai videos about trump fights with zelenskyy. https://vt.tiktok.com/ZSMC4jW6g/ , 2025. Ac- cessed: 2025-03-07. 1

    2025

  65. [73]

    Attention is all you need

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszko- reit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need. NeurIPS, 30, 2017. 5

    2017

  66. [74]

    Pixel-wise crowd understanding via synthetic data

    Qi Wang, Junyu Gao, Wei Lin, and Yuan Yuan. Pixel-wise crowd understanding via synthetic data. IJCV, 129(1):225– 245, 2021. 4

    2021

  67. [75]

    Noise based deepfake de- tection via multi-head relative-interaction

    Tianyi Wang and Kam Pui Chow. Noise based deepfake de- tection via multi-head relative-interaction. In AAAI, pages 14548–14556, 2023. 2, 3, 7, 8

    2023

  68. [76]

    Gender and age classification of human faces for automatic detection of anomalous human behaviour

    Xiaofeng Wang, Azliza Mohd Ali, and Plamen Angelov. Gender and age classification of human faces for automatic detection of anomalous human behaviour. In CYBCONF, pages 1–6, 2017. 4

    2017

  69. [77]

    Deepfake on face and expression swap: A review.IEEE Access, 11:117865–117906, 2023

    Saima Waseem, Syed Abdul Rahman Syed Abu Bakar, Bi- lal Ashfaq Ahmed, Zaid Omar, and Taiseer Abdalla Elfadil Eisa. Deepfake on face and expression swap: A review.IEEE Access, 11:117865–117906, 2023. 1

    2023

  70. [78]

    Active factor graph network for group activity recog- nition

    Zhao Xie, Chang Jiao, Kewei Wu, Dan Guo, and Richange Hong. Active factor graph network for group activity recog- nition. TIP, 2024. 3

    2024

  71. [79]

    Tall: Thumbnail layout for deepfake video detection

    Yuting Xu, Jian Liang, Gengyun Jia, Ziming Yang, Yanhao Zhang, and Ran He. Tall: Thumbnail layout for deepfake video detection. In ICCV, pages 22658–22668, 2023. 2, 3, 7

    2023

  72. [80]

    Spatio-temporal dynamic inference network for group activity recognition

    Hangjie Yuan, Dong Ni, and Mang Wang. Spatio-temporal dynamic inference network for group activity recognition. In ICCV, pages 7476–7485, 2021. 3, 5

    2021

  73. [81]

    Comics: End-to-end bi-grained contrastive learn- ing for multi-face forgery detection

    Cong Zhang, Honggang Qi, Shuhui Wang, Yuezun Li, and Siwei Lyu. Comics: End-to-end bi-grained contrastive learn- ing for multi-face forgery detection. TCSVT, 2024. 2, 3, 4, 7, 8

    2024

  74. [82]

    Gazeonce: Real- time multi-person gaze estimation

    Mingfang Zhang, Yunfei Liu, and Feng Lu. Gazeonce: Real- time multi-person gaze estimation. In CVPR, pages 4197– 4206, 2022. 4

    2022

  75. [83]

    Two-stream neural networks for tampered face detection

    Peng Zhou, Xintong Han, Vlad I Morariu, and Larry S Davis. Two-stream neural networks for tampered face detection. In CVPRW, pages 1831–1839. IEEE, 2017. 4

    2017

  76. [84]

    Face forensics in the wild

    Tianfei Zhou, Wenguan Wang, Zhiyuan Liang, and Jianbing Shen. Face forensics in the wild. In CVPR, pages 5778– 5788, 2021. 2, 3, 4, 6, 7, 8

    2021

  77. [85]

    Muggle: Multi-stream group gaze learning and estimation

    Ning Zhuang, Bingbing Ni, Yi Xu, Xiaokang Yang, Wenjun Zhang, Zefan Li, and Wen Gao. Muggle: Multi-stream group gaze learning and estimation. TCSVT, 30(10):3637–3650,

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.