Pith. sign in

REVIEW 4 major objections 5 minor 26 references

Talking-face deepfakes are detectable from the physiological pulse signal alone, nearly matching state-of-the-art general detectors.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

A classifier trained on rPPG waveforms from face videos detects talking-face deepfakes with AUC 0.806, and detection difficulty varies by generator (AUC 0.690–0.985).

T0 review reviewed 2026-08-01 challenge →

load-bearing objection The aggregate rPPG result and the DeepFakesON-Phys reproduction are solid; the paper's advertised main claim—a stable per-generator physiological ranking—rests on positional method labels the authors admit cannot be verified. the 4 major comments →

arxiv 2607.21776 v1 pith:HHLHNTVY submitted 2026-07-23 cs.LG cs.CRcs.CVcs.MM

Physiological Signals as a Forensic Modality for Talking-Face Deepfake Detection

Othmane Harraq , Tamer Aldwairi This is my paper
classification cs.LG cs.CRcs.CVcs.MM
keywords talking-face deepfake detectionremote photoplethysmography (rPPG)1D ResNetsubject-independent evaluationCeleb-DF++physiological signal forensicsmethod-dependent detectabilityRhythmFormer
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The reading

This paper claims that talking-face deepfakes—videos synthesized from a single still image and an audio track—carry no real physiological substrate, so any apparent heart-pulse signal is an artifact of the generator. The authors therefore propose detecting them by extracting a per-video rPPG waveform from facial skin-color changes and classifying it with a small 1D residual network. On the talking-face subset of Celeb-DF++, evaluated with strict separation of celebrity identities between training and test, the network reaches an AUC of 0.806 (EER 27.8%), within 2.4 points of the best published general-purpose detector while using only the physiological channel. The paper's central discovery is that detection difficulty varies sharply and stably with the generation method—AUC ranges from 0.985 for a 3D-rendering method down to 0.690 for a landmark-warping method—and that this stable ordering reflects an interpretable property of each generator's pixel synthesis strategy rather than evaluation noise.

Core claim

The primary claim is that rPPG-based detection is not just applicable but especially well-motivated for talking-face forgeries, because unlike face-swapping, talking-face synthesis leaves no real video whose blood-flow characteristics could be inherited. The authors show that a 165K-parameter 1D ResNet trained on RhythmFormer pulse waveforms outperforms a reproduced prior rPPG detector (AUC 0.622) by 0.184 points and comes within 2.4 points of the best general-purpose detector on the benchmark. The theoretically significant finding is the per-method AUC spread: Real3DPortrait (0.985), EDTalk (0.950), SadTalker (0.946), AniTalker (0.908), EchoMimic (0.865), FLOAT (0.794), and IP-LAP (0.690),

What carries the argument

The central object is the rPPG (remote photoplethysmography) waveform: a 160-sample cardiac-pulse trace recovered from subtle periodic color variations in facial skin. The pipeline detects faces with a lightweight detector, extracts the waveform with RhythmFormer (a transformer with periodic sparse attention), z-score normalizes each waveform to remove identity-specific amplitude, and feeds it to a 1D ResNet (165K parameters). The per-method analysis treats each generator's output waveform as its physiological signature, using AUC to quantify how well the waveform discriminates real from fake.

Load-bearing premise

The per-method AUC ranking rests on an unverifiable label assignment: because Celeb-DF++ filenames do not carry method names, the authors map each fake video to a generator by its position in the dataset's generation order; if that positional mapping is wrong, the central per-generator ranking collapses, even though the aggregate real-vs-fake detection might survive.

What would settle it

Obtain ground-truth generator labels for the Celeb-DF++ talking-face videos (for example, by querying the benchmark's original metadata or re-synthesizing videos with known method names) and recompute per-method AUC. If the ordering Real3DPortrait > EDTalk > SadTalker > AniTalker > EchoMimic > FLOAT > IP-LAP does not reproduce under verified labels, the claim that the ranking is an interpretable physiological property of each generator is refuted.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • If the claims are correct, rPPG provides a detection channel for talking-face forgeries that does not depend on audio, texture, or frequency features, so it works even when audio is missing or tampered with.
  • The stable per-method ranking implies that the forensic difficulty of a future talking-face generator can be anticipated from its synthesis strategy (latent-pixel synthesis vs source-frame warping) without running detection experiments.
  • The DeepFakesON-Phys reproduction (0.999 on face-swap data, 0.622 on talking-face data) shows that existing rPPG detectors do not transfer to talking-face forgeries; talking-face-specific training is required.
  • IP-LAP's resistance, including the inversion under isolated training, identifies identity-preserving landmark-based synthesis as the main open challenge for physiological forensics.
  • The near-parity with general-purpose detectors on a strict identity-disjoint protocol suggests physiological signals generalize across identities, strengthening the case for including rPPG in multi-modal detectors.

Where Pith is reading between the lines

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

  • A natural extension the paper leaves implicit: the generator-specific waveform shape (incoherent noise vs near-sinusoidal artifact) could be used for generator attribution or fingerprinting, not just binary real/fake classification, since the physiological artifact encodes the synthesis family.
  • The paper's dual-use caveat can be pushed further: if adversarial synthesis injects a plausible cardiac waveform into latent-synthesis methods, the detection gap for those methods may widen; a direct experiment would be to add synthetic rPPG to Real3DPortrait outputs and measure the AUC drop.
  • The acknowledged skin-tone and lighting sensitivity suggests the reported aggregate AUC may not hold uniformly across demographic groups; a stratified evaluation by Fitzpatrick skin type is a testable next step that could revise the per-method ordering for some generators.
  • The authors' reliance on positional method labels for the per-method analysis is a verification gap independent of the aggregate result; if the Celeb-DF++ ordering is ever corrected, the ranking might shift, and the 'physiological fingerprint' interpretation should be re-tested with verified labels.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper proposes the first rPPG-based detection framework specifically for talking-face (TF) deepfakes. It extracts 160-sample pulse waveforms with RhythmFormer from face crops and trains lightweight 1D classifiers (ResNet/CNN/Transformer) under a subject-independent identity split on the TF subset of Celeb-DF++. The authors report AUC 0.806 with the 1D ResNet, reproduce DeepFakesON-Phys at AUC 0.622 on the same TF data, and present a method-stratified analysis showing a per-generator AUC spread from 0.690 (IP-LAP) to 0.985 (Real3DPortrait), interpreted as a stable, generator-specific physiological property linked to each method's pixel synthesis strategy.

Significance. If the main claims hold, the paper makes a useful empirical contribution: a physiological-channel detector for a forgery category where image-based detectors are known to be weak, a controlled reproduction of a prior rPPG detector demonstrating distribution shift, and an analysis suggesting that TF generators differ greatly in rPPG detectability. The paper also ships public code, uses identity-disjoint splits, and reports multi-seed means, which are strengths. However, the paper's advertised primary theoretical contribution—the method-dependent detectability ranking—rests on an explicitly unverifiable positional assignment of method labels, and the headline comparison to Effort is made against a different evaluation protocol. The aggregate real-vs-fake result and the reproduction study are more robust, but the central advertised novelty is not adequately supported as written.

major comments (4)
  1. [§6.4, Table 4, Fig. 2] The primary theoretical contribution—the stable per-generator AUC ranking (0.690–0.985)—depends entirely on the positional assignment of method labels to fake videos. Section 6.4 admits that this assignment 'cannot be independently verified at the video level' because method names are not recoverable from filenames. If the assumed Celeb-DF++ ordering is even partially wrong, Table 4, Fig. 2, Fig. 3, and the interpretation in §6.3 would all be invalid, even though the aggregate real-vs-fake AUC would still stand. The authors must either provide an independently verifiable per-video label mapping (e.g., a released label file tied to dataset metadata) or remove/reframe the per-method claim. This is a load-bearing limitation, not a cosmetic one.
  2. [§5.3, §6.1, §2.5] The headline 'within 2.4 points of Effort (ICML 2025)' compares the proposed TF-only AUC (0.806) with Effort's 0.830 reported on the full Celeb-DF++ benchmark across all forgery categories. Since §2.5 states that TF is the hardest category for existing detectors (AUC 50–70%), this is not a same-protocol comparison and the 'within 2.4 points' claim is not established. The authors should report Effort on the same TF-only split if possible, or explicitly state that no such per-category result exists and soften the claim accordingly.
  3. [§5.4, Table 4, §6.3] The claim that the method ranking is 'perfectly consistent across every evaluation protocol' is not backed by a protocol-by-protocol table; only the 18-identity result is shown. In addition, the mechanistic explanation that detection difficulty is determined by 'pixel synthesis strategy' (warp vs. synthesize) is a post hoc categorization: no direct measure of pixel preservation or temporal color coherence is provided, and the paper does not test this explanation against alternative factors (e.g., video length, compression, identity distribution across methods). The stability claim and the causal-sounding interpretation should be either directly supported with the full AUC matrix and quantitative covariates, or explicitly downgraded to a hypothesis.
  4. [§3.1, §4.3, §5.1] The evaluation unit is ambiguous. Section 3.1 defines the classifier input as a 'per-video rPPG waveform', but §4.3 windows long videos into multiple 160-frame clips at stride 60, and Table 3 reports AUC without stating whether it is window-level or video-level. The DeepFakesON-Phys reproduction in §5.1, by contrast, is explicitly 'video-level'. If the main AUC numbers are window-level, they are not directly comparable to video-level benchmarks and may benefit from within-video autocorrelation. The authors should specify the unit of evaluation, aggregate to video-level if that is the forensic claim, or report both.
minor comments (5)
  1. [Abstract, §4.1] The text says 'yielding 585 real waveforms (585 source videos with stride-60 temporal windowing produce 2,435 waveforms in total)' and then reports 2,435 real training waveforms. The '585 real waveforms' phrase appears to be a typo and should be corrected for clarity.
  2. [§4.2, §5.3] Table 1 lists a 9-identity validation split, but the main result in §5.3 uses an '18-identity test set (val and test identities combined)'. Clarify whether the validation identities were used in any model-selection or early-stopping procedure before being combined into the final evaluation; if they were fully untouched, state this explicitly.
  3. [Fig. 3] The waveforms shown are explicitly 'representative' and appear to have been selected after seeing the results. This is acceptable as illustration, but the caption should state that these are hand-picked examples, not a systematic characterization of each method's output.
  4. [§2.5] The statement that 'robust results are limited to FS forgeries (AUC>85%), while FR and TF forgeries yield degraded detection rates (AUC 50–70%)' is not followed by a citation or a table. Please add the source of these numbers.
  5. [§5.4] The term 'theoretical contribution' is used in the abstract and §5.4 for what is an empirical, method-stratified observation. Consider using 'empirical finding' or 'forensic characterization' unless the paper provides a formal theoretical model predicting the ranking.

Circularity Check

0 steps flagged

No significant circularity: the detection pipeline is fit on held-out identities, and the per-method interpretability is post hoc explanation, not a fitted input.

full rationale

The claimed derivation chain is a standard supervised evaluation, not a circular reduction. RhythmFormer is a pretrained external rPPG extractor (UBFC checkpoint), and the 1D ResNet is trained on subject-disjoint splits; the target label (real vs. TF fake) is never used to fit the waveform extractor or to define the generator ranking. The aggregate AUC 0.806 is therefore an independent measurement, and the comparison against Effort is an external benchmark. The per-method AUC spread (0.690-0.985) is a descriptive stratification of the same held-out test set; the explanation in terms of pixel synthesis strategy is a post hoc qualitative interpretation that does not set any parameter, appear in the loss, or enter the ROC computation. The paper's own acknowledged limitation (Section 6.4, "method names are not recoverable from filenames... cannot be independently verified at the video level") concerns the provenance of method labels, not a circular fit; a mislabeled input would be a correctness risk, not a self-referential derivation. There are no self-citations, no imported uniqueness theorem, and no ansatz smuggled in by citation: RhythmFormer and DeepFakesON-Phys are external prior work with released weights. Thus no prediction in the paper reduces by construction to its inputs, and the circularity score is 0.

Axiom & Free-Parameter Ledger

11 free parameters · 5 axioms · 0 invented entities

The central empirical result (AUC 0.806) rests on standard dataset and model-transfer assumptions; the more novel per-method claim rests on an ad hoc, unverifiable positional label assignment and a post-hoc interpretation of synthesis strategy. The hyperparameter choices are standard ML tuning and are listed as free parameters.

free parameters (11)
  • 1D ResNet learning rate = 1e-3
    Tuned via 5-fold cross-validation (Sec. 3.4, 4.4); affects final AUC.
  • 1D ResNet weight decay = 5e-4
    Tuned via cross-validation; affects final AUC.
  • 1D ResNet dropout = 0.5
    Tuned via cross-validation; affects generalization.
  • Transformer learning rate = 5e-4
    Tuned via cross-validation (Sec. 3.4).
  • Transformer weight decay = 1e-3
    Tuned via cross-validation.
  • Transformer dropout = 0.1
    Tuned via cross-validation.
  • Temporal window length = 160 frames
    Hand-chosen; defines the waveform length input to classifiers (Sec. 3.2).
  • Temporal stride = 60 frames
    Hand-chosen for training real-video augmentation (Sec. 4.3); affects number of waveforms.
  • Face bounding-box expansion = 1.5x
    Hand-chosen context margin for face crops (Sec. 3.2).
  • Zero-mask length (augmentation) = U[10,25] frames
    Hand-chosen augmentation distribution (Sec. 3.5).
  • Z-score epsilon = 1e-6
    Hand-chosen to avoid division by zero (Eq. 2).
axioms (5)
  • domain assumption RhythmFormer pretrained on UBFC-rPPG transfers to Celeb-DF++ faces and produces waveforms that carry physiological information.
    Section 3.2 loads a UBFC checkpoint; no validation against ground-truth pulse on Celeb-DF++ is provided.
  • domain assumption Celeb-DF++ TF subset real/fake labels are correct.
    Section 4.1; the paper relies on the benchmark without independent verification.
  • ad hoc to paper Positional assignment of method labels to fake videos matches the actual generator for each video.
    Section 6.4 admits this cannot be independently verified at the video level.
  • domain assumption z-score normalization of each waveform removes identity-specific baseline while preserving discriminative temporal structure.
    Equation (2); Table 2 shows it has little effect on AUC, so the assumption is used to justify the representation.
  • ad hoc to paper Per-generator detection difficulty is determined by pixel synthesis strategy (warp vs. synthesize).
    Section 5.4 and 6.3; asserted post hoc without an independent measure of synthesis strategy.

reviewed 2026-08-01 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Physiological Signals as a Forensic Modality for Talking-Face Deepfake Detection." pith.science (2026). https://pith.science/paper/HHLHNTVY

@misc{pith2026260721776,
  author       = {Pith},
  title        = {Pith review of: Physiological Signals as a Forensic Modality for Talking-Face Deepfake Detection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HHLHNTVY}},
  note         = {Machine review of arXiv:2607.21776}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Talking-face (TF) deepfake generation synthesizes photore- alistic facial video from a static source image and an au- dio signal, producing forgeries that current image-based detectors consistently fail to identify. Unlike face-swap ma- nipulation, TF synthesis has no underlying real video from which to inherit physiological characteristics, making re- mote photoplethysmography (rPPG) a uniquely motivated detection modality for this forgery category. We propose a detection framework that extracts per-video rPPG wave- forms via RhythmFormer and trains a suite of lightweight classifiers to distinguish real from synthesized physiologi- cal signals. Evaluated on the TF subset of Celeb-DF++ un- der a strict subject-independent protocol, where test identi- ties are completely separated from training identities, our 1D ResNet achieves an AUC of 0.806 and EER of 27.8%, placing it within 2.4 points of the best published general- purpose detector (Effort, ICML 2025) while operating ex- clusively on the physiological channel. We document a con- trolled reproduction study of DeepFakesON-Phys, the rep- resentative prior rPPG detector, demonstrating degrada- tion from AUC 0.999 on legacy face-swap data to 0.622 on the TF subset of Celeb-DF++. We further show that detec- tion difficulty is strongly method-dependent: AUC ranges from 0.985 (Real3DPortrait) to 0.690 (IP-LAP) across the seven TF generators, with the ranking remaining perfectly stable across all evaluation protocols. This spread reflects an interpretable physiological property of each generator rather than evaluation noise, and constitutes the primary theoretical contribution of the work.

Figures

Figures reproduced from arXiv: 2607.21776 by Othmane Harraq, Tamer Aldwairi.

Figure 1
Figure 1. Figure 1: Overview of the proposed rPPG-based TalkingFace de [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: ROC curves for the 1D ResNet evaluated per Talking [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Representative rPPG waveforms (z-score normalized) [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

26 extracted references · 7 linked inside Pith

  1. [1]

    Deepfake video detection through optical flow based CNN

    Giuseppe Boccignone et al. Deepfake video detection through optical flow based CNN. InInt. Conf. Pattern Recog., 2022. 1

    2022

  2. [2]

    EchoMimic: Lifelike audio-driven portrait animations through editable landmark conditions

    Zhiyuan Chen et al. EchoMimic: Lifelike audio-driven portrait animations through editable landmark conditions. arXiv:2407.08136, 2024. 1, 2

    Pith/arXiv arXiv 2024

  3. [3]

    Fake- Catcher: Detection of synthetic portrait videos using bio- logical signals.IEEE Transactions on Pattern Analysis and Machine Intelligence, 2020

    Umur Aybars Ciftci, Ilke Demir, and Lijun Yin. Fake- Catcher: Detection of synthetic portrait videos using bio- logical signals.IEEE Transactions on Pattern Analysis and Machine Intelligence, 2020. 1, 2

    2020

  4. [4]

    Detecting lip-syncing deepfakes: Vision temporal transformer for ana- lyzing mouth inconsistencies.arXiv:2504.01470, 2025

    Soumyya Kanti Datta, Shan Jia, and Siwei Lyu. Detecting lip-syncing deepfakes: Vision temporal transformer for ana- lyzing mouth inconsistencies.arXiv:2504.01470, 2025. 2

    Pith/arXiv arXiv 2025

  5. [5]

    Robust pulse rate from chrominance-based rPPG.IEEE Transactions on Biomedical Engineering, 60(10):2878–2886, 2013

    Gerard de Haan and Vincent Jeanne. Robust pulse rate from chrominance-based rPPG.IEEE Transactions on Biomedical Engineering, 60(10):2878–2886, 2013. 2

    2013

  6. [6]

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

    Alexandros Haliassos et al. Lips don’t lie: A generalisable and robust approach to face forge detection. InIEEE Conf. Comput. Vis. Pattern Recog., 2021. 1, 2

    2021

  7. [7]

    DeepFakesON-Phys: DeepFakes de- tection based on heart rate estimation.arXiv:2010.00400,

    Javier Hernandez-Ortega, Ruben Tolosana, Julian Fierrez, and Aythami Morales. DeepFakesON-Phys: DeepFakes de- tection based on heart rate estimation.arXiv:2010.00400,

    Pith/arXiv arXiv 2010

  8. [8]

    FLOAT: Generative motion latent flow matching for audio-driven talking portrait

    Taekyung Ki, Dongchan Min, and Gyeongsu Chae. FLOAT: Generative motion latent flow matching for audio-driven talking portrait. InInt. Conf. Comput. Vis., 2025. 1, 2

    2025

  9. [9]

    Celeb- DF++: A large-scale challenging video deepfake benchmark for generalizable forensics.arXiv:2507.18015, 2025

    Yuezun Li, Delong Zhu, Xinjie Cui, and Siwei Lyu. Celeb- DF++: A large-scale challenging video deepfake benchmark for generalizable forensics.arXiv:2507.18015, 2025. 1, 2, 4

    Pith/arXiv arXiv 2025

  10. [10]

    Identifying deepfake videos with optical flow and biological signals.arXiv:2106.11468,

    Tianyi Liang and Xin Deng. Identifying deepfake videos with optical flow and biological signals.arXiv:2106.11468,

    Pith/arXiv arXiv

  11. [11]

    AniTalker: Animate vivid and diverse talk- ing faces via identity-decoupled facial motion encoding

    Tao Liu et al. AniTalker: Animate vivid and diverse talk- ing faces via identity-decoupled facial motion encoding. arXiv:2405.03121, 2024. 1, 2

    Pith/arXiv arXiv 2024

  12. [12]

    Lips are lying: Spotting the temporal incon- sistency between audio and visual in lip-syncing deepfakes

    Xiaoyi Liu et al. Lips are lying: Spotting the temporal incon- sistency between audio and visual in lip-syncing deepfakes. arXiv:2401.15668, 2024. 1, 2

    Pith/arXiv arXiv 2024

  13. [13]

    A meta-analysis of the impact of skin tone and gender on non-contact photoplethysmography measurements

    Ewa M Nowara, Daniel McDuff, and Ashok Veeraraghavan. A meta-analysis of the impact of skin tone and gender on non-contact photoplethysmography measurements. InIEEE Conf. Comput. Vis. Pattern Recog. Worksh., 2020. 7

    2020

  14. [14]

    SpecAugment: A simple data augmen- tation method for automatic speech recognition

    Daniel S Park et al. SpecAugment: A simple data augmen- tation method for automatic speech recognition. InInter- speech, 2019. 4

    2019

  15. [15]

    DeepRhythm: Exposing deepfakes with attentional visual heartbeat rhythms

    Hanqing Qi et al. DeepRhythm: Exposing deepfakes with attentional visual heartbeat rhythms. InACM Int. Conf. Mul- timedia, 2020. 1, 2

    2020

  16. [16]

    EDTalk: Efficient disentanglement for emo- tional talking head synthesis

    Shuai Tan et al. EDTalk: Efficient disentanglement for emo- tional talking head synthesis. InEur. Conf. Comput. Vis.,

  17. [17]

    Re- mote plethysmographic imaging using ambient light.Optics Express, 16(26):21434–21445, 2008

    Wim Verkruysse, Lars O Svaasand, and J Stuart Nelson. Re- mote plethysmographic imaging using ambient light.Optics Express, 16(26):21434–21445, 2008. 1, 2

    2008

  18. [18]

    Algorithmic principles of remote PPG

    Wenjin Wang, Albertus C den Brinker, Sander Stuijk, and Gerard de Haan. Algorithmic principles of remote PPG. IEEE Transactions on Biomedical Engineering, 64(7):1479– 1491, 2017. 1, 2

    2017

  19. [19]

    Detecting deepfake videos with temporal dropout 3DCNN

    Yue Wu et al. Detecting deepfake videos with temporal dropout 3DCNN. InIJCAI, 2023. 1, 2

    2023

  20. [20]

    Orthogonal subspace decom- position for generalizable AI-generated image detection

    Zhiyuan Yan, Jiangming Wang, Peng Jin, Ke-Yue Zhang, Chengchun Liu, Shen Chen, Taiping Yao, Shouhong Ding, Baoyuan Wu, and Li Yuan. Orthogonal subspace decom- position for generalizable AI-generated image detection. In Proceedings of the 42nd International Conference on Ma- chine Learning, pages 70268–70288. PMLR, 2025. 2, 5

    2025

  21. [21]

    Real3D-Portrait: One-shot realistic 3D talking portrait synthesis

    Zhenhui Ye et al. Real3D-Portrait: One-shot realistic 3D talking portrait synthesis. InInt. Conf. Learn. Represent.,

  22. [22]

    Remote photoplethysmograph signal mea- surement from facial videos using spatio-temporal networks

    Zitong Yu et al. Remote photoplethysmograph signal mea- surement from facial videos using spatio-temporal networks. InBrit. Mach. Vis. Conf., 2019. 2

    2019

  23. [23]

    PhysFormer++: Facial video-based phys- iological measurement with slowfast temporal difference transformer.Int

    Zitong Yu et al. PhysFormer++: Facial video-based phys- iological measurement with slowfast temporal difference transformer.Int. J. Comput. Vis., 2023. 2

    2023

  24. [24]

    SadTalker: Learning realistic 3D mo- tion coefficients for stylized audio-driven single image talk- ing face animation

    Wenxuan Zhang et al. SadTalker: Learning realistic 3D mo- tion coefficients for stylized audio-driven single image talk- ing face animation. InIEEE Conf. Comput. Vis. Pattern Recog., 2023. 1, 2

    2023

  25. [25]

    Identity-preserving talking face gen- eration with landmark and appearance priors

    Zhimeng Zhong et al. Identity-preserving talking face gen- eration with landmark and appearance priors. InIEEE Conf. Comput. Vis. Pattern Recog., 2023. 1, 2

    2023

  26. [26]

    RhythmFormer: Extracting patterned rPPG signals based on periodic sparse attention.Pattern Recognition, 164, 2025

    Bochao Zou, Zizheng Guo, Jiansheng Chen, and Huimin Ma. RhythmFormer: Extracting patterned rPPG signals based on periodic sparse attention.Pattern Recognition, 164, 2025. 1, 2, 3 8

    2025

This paper was first reviewed by deepseek-v4-flash on August 1, 2026.