Pith. sign in

REVIEW 3 major objections 6 minor 66 references

Test-Time Augmentation for Pose-invariant Face Recognition

T0 review · 3 major / 6 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read Pose-TTA claims that a pre-trained face recognition model can be improved at inference, without retraining, by generating a matching side-profile view of the more frontal face and aggregating the two embeddings with fixed weights.

desk verdict Pose-TTA is a modest, clearly explained inference-time plug-in that improves pose-heavy face verification with portrait animation; the broad experiments are useful, but the 'consistently improves' claim overstates the tables and identity preservation is never directly measured. read the letter →

arxiv 2505.09256 v1 pith:3I5AYUNW submitted 2025-05-14 cs.CV

Jaemin Jung , Youngjoon Jang , Joon Son Chung This is my paper · ORCID
classification cs.CV
keywords test-timeaugmentationfacerecognitionpose-invariantportraitanimationfeatureaggregationheadposealignmentverificationfrontalisation
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

Pose-TTA claims that a frozen face recognition model can be made more accurate on pose-diverse image pairs at inference time, with no retraining or fine-tuning. The idea is to stop trying to frontalise side-profile faces and instead use a portrait animator to generate a matching side-profile view from the more frontal image, then compare the two aligned views. A small face selector chooses which image supplies identity and which supplies head pose, and a weighted average of the original and synthetic embeddings (weight 0.75 on real, 0.25 on synthetic) keeps generative artifacts from dominating. On pose-heavy benchmarks the method reports average gains of about 0.72 percentage points on CPLFW and 0.19 on CFP-FP across several pre-trained backbones and training sets, while leaving near-frontal benchmarks roughly unchanged. If correct, this gives a plug-in inference wrapper that improves pose robustness without dataset-specific training.

What carries the argument

The central mechanism is the source/driving split with symmetry-guided flipping. The face with the smaller absolute yaw is treated as the identity source; the face with the larger absolute yaw contributes only its head pose. When the yaw directions disagree, the source is horizontally flipped before being fed to the portrait animator, so the two faces point roughly the same way and the generator does not have to hallucinate the occluded half of the face. The second load-bearing piece is the fixed-weight aggregation: synthetic embeddings are down-weighted ($w_{syn}=0.25$, $w_{real}=0.75$) relative to the real image, and the combined vector is normalised along the channel axis before verification. Together these pieces convert test-time augmentation from a generic averaging step into a pose-alignment step that preserves identity information.

What would settle it

Take a pose-heavy benchmark such as CPLFW and replace every driving image's identity with a different person's face while keeping the yaw angle; if the reported gain persists, pose alignment is doing the work, whereas if it disappears, the animator is injecting driving-identity information into the augmented embeddings.

Watch

Extended reading notes

Core claim

On its own terms, the paper's claim is that pose-invariant face recognition can be decomposed at test time into two steps: generate, rather than reconstruct. Given a pair of faces, the face selector uses an off-the-shelf head pose estimator to label the image with smaller absolute yaw as the source (identity) and the image with larger absolute yaw as the driving image (pose only). If the two yaw signs differ, the source is horizontally flipped so the animator transfers pose without having to invent the unseen side of the face. The animator copies only head pose, not expression, so identity-related details such as eye and mouth shape are not biased toward the driving person. Both the original and the generated images, plus their horizontal flips, are embedded by the frozen recognition model and combined with fixed weights $\hat{\mathbf{y}}_{tta} = \frac{1}{|T|}\sum_i w_T \Theta_{target}(\tau_i(x))$, where $w_{real}=0.75$ and $w_{syn}=0.25$, followed by channel-wise normalisation. The reported result is that this pipeline consistently improves accuracy on pose-heavy benchmarks while barely changing frontal-dominant ones.

Load-bearing premise

The method assumes that the portrait animator, after the optional horizontal flip, preserves identity closely enough that the generated side-profile embedding behaves like another view of the same person.

Editorial extensions

If this is right

  • A pre-trained face recognition model improves on pose-diverse pairs (average +0.72 on CPLFW, +0.19 on CFP-FP) without any training or fine-tuning.
  • Frontalisation at test time is the wrong comparison: frontalising the profile hurts pose-heavy accuracy (about -0.12 average), while matching side profiles gains about +0.37 average.
  • The weighting is doing real work: equal weights (0.5/0.5) underperform 0.75/0.25, and relying entirely on synthetic images drops CPLFW accuracy from 87.87 to 85.77.
  • The horizontal-flip alignment contributes a small but consistent gain (about +0.04 average), so facial symmetry is part of the method, not decoration.
  • Because the recognition model is frozen, the wrapper can be attached to different backbones (ResNet18/50/101, ViT) and training losses (AdaFace, ArcFace) without per-dataset retraining.

Reading between the lines

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

  • A natural test of the mechanism is to scramble the driving image's identity: if pose alignment alone drives the gain, the reported improvement should survive; if it collapses, the animator is leaking identity from the driving face and the aggregation is mainly masking that leakage.
  • The fixed 0.75/0.25 weights are tuned on these benchmarks; an input-dependent reliability estimate (for instance, from the animator's own stitching confidence) could recover part of the small losses seen on frontal-heavy sets such as AgeDB-30 (-0.04).
  • The wrapper nature of the method means it could be stacked with future encoders or pose estimators; the only moving parts that need updating are the animator and the two weights.
  • A direct identity-preservation audit of the generated profiles (does the synthetic face retrieve the right person in a large gallery?) would tell whether the gains are pose alignment or synthetic-data artifacts.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. The paper introduces Pose-TTA, a test-time augmentation method for face recognition. For a pair of face images, a Face Selector uses an off-the-shelf yaw estimator to designate the more frontal image as the source and the more profile image as the driving image; LivePortrait animates the source into the driving pose, with an optional horizontal flip of the source when the yaw directions disagree. The pre-trained recognition model's embeddings for the original, flipped, synthetic, and flipped-synthetic images are aggregated with fixed weights wreal=0.75, wsyn=0.25 and L2-normalized before verification. Experiments cover five benchmarks (CPLFW, CFP-FP, LFW, CALFW, AgeDB), several backbones/losses, and training sets including the synthetic DCFace set. The paper claims consistent improvements, particularly on pose-heavy benchmarks, and compares against frontalisation and unweighted averaging.

Significance. The idea of generating matching side-profile views rather than frontalising is interesting and practically attractive because it requires no retraining. The evaluation across six training sets and five architectures is a strength, and Table II provides a useful comparison against frontalisation. However, the central claim is only partially supported: Table I contains multiple negative deltas, the gains are concentrated on CPLFW and CFP-FP, and no statistical significance is reported. The method also depends on two hyperparameters tuned on the same test benchmarks, and the identity-preservation premise of the portrait animator is not measured. If these issues are addressed, the contribution could be a useful incremental result for inference-time pose robustness.

major comments (3)
  1. [Section III-C, Table III] The weights wreal=0.75 and wsyn=0.25 are selected by evaluating the five test benchmarks in Table III and then used to report the final results on those same benchmarks in Tables I and II. This is test-set tuning; it makes the reported gains optimistic and weakens the claim that the method is a fixed, training-free procedure. Please either select weights on a validation split or via cross-validation, or report results across the weight grid and clearly state that the reported configuration is the best on the test sets.
  2. [Section II-A, Eq. (2), Fig. 2] The paper's core premise is that LivePortrait preserves the source identity while transferring pose, but no quantitative identity-preservation measurement is provided; Figure 2 is qualitative. If the animator transfers identity cues from the driving image, the synthetic embedding used in Eq. (2) can inflate impostor similarity, and the observed gains may not reflect pose alignment. Please add a quantitative identity-retention evaluation, for example embedding similarity of the generated image to the source versus to the driving image, or verification accuracy on synthetic source/driving pairs, using the same pre-trained recognition model.
  3. [Abstract, Section IV-A, Table I] The statement that Pose-TTA 'consistently improves inference performance' is contradicted by Table I, which contains negative deltas on CFP-FP, LFW, CALFW, and AgeDB (for example AdaFace ResNet50 WebFace4M has -0.10 on CFP-FP, and multiple LFW and AgeDB entries are negative). The later text acknowledges marginal drops, which should be reconciled with the abstract. Moreover, no confidence intervals or significance tests are provided, so gains of +0.05 or +0.09 cannot be distinguished from noise. Please revise the claims to match the evidence and add appropriate statistical evaluation.
minor comments (6)
  1. [Section II-A] The statement that the head pose estimator 'is not used during the pose augmentation process, thereby eliminating dependence on its performance' is inaccurate because the Face Selector uses yaw values to choose the source/driving images and to decide the horizontal flip; please revise the wording.
  2. [Table I caption] The caption says that baseline results 'without TTA' use the aggregation of original and flipped embeddings, but that is itself a form of test-time augmentation; please clarify that the baseline is flip-only TTA rather than 'without TTA'.
  3. [Table I, Fig. 1] There are several proofreading issues: the average accuracy '94,97' uses a comma as a decimal separator, Fig. 1 contains the typo 'recongition', and some delta entries in Table I do not match the difference between the printed baseline and TTA accuracy columns (for example AgeDB row 1, where 98.00 to 97.87 is -0.13 but the table prints -0.03).
  4. [Ethical Impact Statement] The Ethical Impact Statement claims that the authors analysed Pose-TTA's performance across demographic groups, but no such subgroup analysis appears in the experiments; please either add the analysis or remove the claim.
  5. [Section II-B] The section is titled 'Test Time Adaptation' although the method is test-time augmentation; please use the terminology consistently.
  6. [Section III-A] Since the method adds a portrait animation step at inference, reporting per-pair runtime or computational cost would help assess practical utility; no timing or cost comparison is currently provided.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation; Pose-TTA is an empirical test-time pipeline validated on external benchmarks.

full rationale

The paper makes no first-principles claim that reduces to its inputs. Its components are external: LivePortrait [18], an off-the-shelf pose estimator [41], and pre-trained face recognition models [11, 27]; the proposed contribution is an inference-time procedure evaluated on standard benchmarks. Equations (1) and (2) are standard test-time averaging with scalar weights, and Eq. (3) only assigns the values wreal=0.75 and wsyn=0.25. No reported accuracy is defined in terms of these equations, and no equation is defined in terms of the reported accuracy. The main caveat is that Table III selects these weights on the same benchmarks later reported in Table I, which is test-set hyperparameter tuning and could overstate gains; however, the paper discloses this as an ablation, and the effect is not forced by construction — with uniform weights (wreal=0.5), Pose-TTA still improves CPLFW by +0.88 points over the baseline, so the central claim does not reduce to the fitted constants. Self-citations [23, 24, 30] are contextual and never load-bearing. The identity-preservation premise is asserted qualitatively but that is an empirical validity concern, not a circularity.

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

The method introduces no new entities. It relies on three external model assumptions (portrait animator, head pose estimator, face recognition model) and two tuned weights. The weights are the only explicit free parameters, and they are fitted to the same benchmarks that are reported as results.

free parameters (2)
  • wreal = 0.75
    Weight for real (original and flipped) embeddings in Eq. (3), selected from ablation on the same evaluation benchmarks (Table III); with wreal=0.5 the average gain is smaller.
  • wsyn = 0.25
    Weight for synthetic augmented embeddings in Eq. (3), paired with wreal=0.75; the ablation shows this combination beats equal weighting, indicating the value is tuned to the target data.
assumptions (4)
  • domain assumption LivePortrait preserves identity of the source face when transferring the driving pose, especially after horizontal flipping.
    In Section II-A, the augmented image is used directly as a face embedding input; if identity leaks toward the driving image, the aggregation is biased.
  • domain assumption Facial symmetry justifies horizontal flipping as a pose alignment operation.
    The flip is applied whenever source and driving yaw have opposite signs (Section II-A); this assumes faces are symmetric enough that flipping does not change identity.
  • domain assumption The head pose estimator's yaw values are accurate enough to pick source and driving images and to decide flip direction.
    Face Selector relies on an off-the-shelf yaw estimator (Section II-A) to determine which image is more frontal and whether to flip.
  • domain assumption Weighted linear combination of face recognition embeddings preserves or improves verification performance.
    Feature aggregation in Section II-B averages real and synthetic embeddings with fixed weights and then normalizes; this assumes the recognition model's embedding space is approximately linear for this purpose.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Test-Time Augmentation for Pose-invariant Face Recognition." pith.science (2026). https://pith.science/paper/3I5AYUNW

@misc{pith2026250509256,
  author       = {Pith},
  title        = {Pith review of: Test-Time Augmentation for Pose-invariant Face Recognition},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3I5AYUNW}},
  note         = {Machine review of arXiv:2505.09256}
}
read the original abstract

The goal of this paper is to enhance face recognition performance by augmenting head poses during the testing phase. Existing methods often rely on training on frontalised images or learning pose-invariant representations, yet both approaches typically require re-training and testing for each dataset, involving a substantial amount of effort. In contrast, this study proposes Pose-TTA, a novel approach that aligns faces at inference time without additional training. To achieve this, we employ a portrait animator that transfers the source image identity into the pose of a driving image. Instead of frontalising a side-profile face -- which can introduce distortion -- Pose-TTA generates matching side-profile images for comparison, thereby reducing identity information loss. Furthermore, we propose a weighted feature aggregation strategy to address any distortions or biases arising from the synthetic data, thus enhancing the reliability of the augmented images. Extensive experiments on diverse datasets and with various pre-trained face recognition models demonstrate that Pose-TTA consistently improves inference performance. Moreover, our method is straightforward to integrate into existing face recognition pipelines, as it requires no retraining or fine-tuning of the underlying recognition models.

Figures

Figures reproduced from arXiv: 2505.09256 by the authors.

Figure 1
Figure 1. An overview of our Pose-TTA framework. In the head pose alignment stage, the source and driving images, [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Head pose augmentation quality comparison. The proposed method [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

66 extracted references · 59 canonical work pages

  1. [1]

    Z. An, W. Deng, Y . Zhong, Y . Huang, and X. Tao. Apa: Adaptive pose alignment for robust face recognition. In Proc. CVPRW , 2019. 1, 3

    work page 2019

  2. [2]

    Atzori, F

    A. Atzori, F. Boutros, N. Damer, G. Fenu, and M. Marras. If it’s not enough, make it so: Reducing authentic data demand in face recognition through synthetic faces. In Proc. FG, 2024. 1

    work page 2024

  3. [3]

    G. Bae, M. de La Gorce, T. Baltru ˇsaitis, C. Hewitt, D. Chen, J. Valentin, R. Cipolla, and J. Shen. Digiface-1m: 1 million digital face images for face recognition. In Proc. WACV, pages 3526–3535,

    work page

  4. [4]

    Banerjee, J

    S. Banerjee, J. Brogan, J. Krizaj, A. Bharati, B. R. Webster, V . Struc, P. J. Flynn, and W. J. Scheirer. To frontalize or not to frontalize: Do we really need elaborate pre-processing to improve face recognition? In Proc. WACV, 2018. 1

    work page 2018

  5. [5]

    Buolamwini and T

    J. Buolamwini and T. Gebru. Gender shades: Intersectional accuracy disparities in commercial gender classification. In Conference on fairness, accountability and transparency , pages 77–91. PMLR, 2018. 5

    work page 2018

  6. [6]

    Q. Cao, L. Shen, W. Xie, O. M. Parkhi, and A. Zisserman. Vggface2: A dataset for recognising faces across pose and age. In Proc. FG ,

    work page

  7. [7]

    Chen and S

    C. Chen and S. McCloskey. High-resolution image enumeration for low-resolution face recognition. In Proc. FG, 2024. 1

    work page 2024

  8. [8]

    S. Chen, Y . Liu, X. Gao, and Z. Han. Mobilefacenets: Efficient cnns for accurate real-time face verification on mobile devices. In Biometric Recognition: 13th Chinese Conference, CCBR 2018, Urumqi, China, August 11-12, 2018, Proceedings 13 , 2018. 1

    work page 2018

Show all 66 references
  1. [9]

    Chu and T

    X. Chu and T. Harada. Generalizable and animatable gaussian head avatar. In Proc. NeurIPS, 2024. 1

    2024

  2. [10]

    J. S. Chung, A. Nagrani, and A. Zisserman. V oxceleb2: Deep speaker recognition. In Proc. Interspeech, 2018. 1

    2018

  3. [11]

    J. Deng, J. Guo, N. Xue, and S. Zafeiriou. Arcface: Additive angular margin loss for deep face recognition. In Proc. CVPR, 2019. 1, 3, 5

    2019

  4. [12]

    J. Deng, J. Guo, D. Zhang, Y . Deng, X. Lu, and S. Shi. Lightweight face recognition challenge. In Proc. CVPRW, 2019. 1, 3, 5

    2019

  5. [13]

    Y . Deng, D. Wang, and B. Wang. Portrait4d-v2: Pseudo multi-view data creates better 4d head synthesizer. In Proc. ECCV, 2024. 1

    2024

  6. [14]

    Dosovitskiy, L

    A. Dosovitskiy, L. Beyer, A. Kolesnikov, D. Weissenborn, X. Zhai, T. Unterthiner, M. Dehghani, M. Minderer, G. Heigold, S. Gelly, J. Uszkoreit, and N. Houlsby. An image is worth 16x16 words: Transformers for image recognition at scale. In Proc. ICLR , 2021. 3

    2021

  7. [15]

    Duan and L

    Q. Duan and L. Zhang. Boostgan for occlusive profile face frontal- ization and recognition. arXiv preprint arXiv:1902.09782 , 2019. 1

    1902 arXiv

  8. [16]

    C. Fu, Y . Hu, X. Wu, G. Wang, Q. Zhang, and R. He. High-fidelity face manipulation with extreme poses and expressions. IEEE Transactions on Information F orensics and Security , 16:2218–2231, 2021. 1

    2021

  9. [17]

    Z. Gao, Q. Li, G. Wang, and L. Shen. Pointfaceformer: local and global attention based transformer for 3d point cloud face recognition. In Proc. FG, 2024. 1

    2024

  10. [18]

    J. Guo, D. Zhang, X. Liu, Z. Zhong, Y . Zhang, P. Wan, and D. Zhang. Liveportrait: Efficient portrait animation with stitching and retargeting control. arXiv preprint arXiv:2407.03168 , 2024. 1, 2

    2024 arXiv

  11. [19]

    Y . Han, J. Zhu, K. He, X. Chen, Y . Ge, W. Li, X. Li, J. Zhang, C. Wang, and Y . Liu. Face-adapter for pre-trained diffusion models with fine-grained id and attribute control. In Proc. ECCV, 2024. 2

    2024

  12. [20]

    K. He, X. Zhang, S. Ren, and J. Sun. Deep residual learning for image recognition. In Proc. CVPR, 2016. 3

    2016

  13. [21]

    Y . Hu, X. Wu, B. Yu, R. He, and Z. Sun. Pose-guided photorealistic face rotation. In Proc. CVPR, 2018. 1

    2018

  14. [22]

    G. B. Huang, M. Mattar, T. Berg, and E. Learned-Miller. Labeled faces in the wild: A database forstudying face recognition in unconstrained environments. In Workshop on faces in‘Real-Life’Images: detection, alignment, and recognition , 2008. 3

    2008

  15. [23]

    Y . Jang, K. Rho, J. Woo, H. Lee, J. Park, Y . Lim, B.-Y . Kim, and J. S. Chung. That’s what i said: Fully-controllable talking face generation. In Proc. ACM MM , 2023. 1

    2023

  16. [24]

    J. Jung, J. Ahn, C. Jung, T. D. Nguyen, Y . Jang, and J. S. Chung. V oicedit: Dual-condition diffusion transformer for environment-aware speech synthesis. In Proc. ICASSP, 2025. 5

    2025

  17. [25]

    I. Kim, Y . Kim, and S. Kim. Learning loss for test-time augmentation. Proc. NeurIPS, 2020. 1

    2020

  18. [26]

    M. Kim. CVLFace: High-Performance Face Recognition All-in-One Toolkit, 2024. 3

    2024

  19. [27]

    M. Kim, A. K. Jain, and X. Liu. Adaface: Quality adaptive margin for face recognition. In Proc. CVPR, 2022. 3

    2022

  20. [28]

    M. Kim, F. Liu, A. Jain, and X. Liu. Dcface: Synthetic face generation with dual condition diffusion model. In Proc. CVPR, 2023. 3, 5

    2023

  21. [29]

    M. Kimura. Understanding test-time augmentation. In Proc. NeurIPS,

  22. [30]

    D. Kwak, J. Jung, K. Nam, Y . Jang, J.-W. Jung, S. Watanabe, and J. S. Chung. V oxmm: Rich transcription of conversations in the wild. In Proc. ICASSP, 2024. 1

    2024

  23. [31]

    D. Leslie. Understanding bias in facial recognition technologies. arXiv preprint arXiv:2010.07023, 2020. 5

    2010 arXiv

  24. [32]

    T. Li, Y . Tian, H. Li, M. Deng, and K. He. Autoregressive image gen- eration without vector quantization. Advances in Neural Information Processing Systems, 37:56424–56445, 2024. 5

    2024

  25. [33]

    W. Liu, Y . Wen, Z. Yu, M. Li, B. Raj, and L. Song. Sphereface: Deep hypersphere embedding for face recognition. In Proc. CVPR, 2017. 1

    2017

  26. [34]

    X. Luan, Z. Ding, L. Liu, W. Li, and X. Gao. A symmetrical siamese network framework with contrastive learning for pose-robust face recognition. IEEE Trans. on Image Processing , 2023. 1

    2023

  27. [35]

    Mehrabi, F

    N. Mehrabi, F. Morstatter, N. Saxena, K. Lerman, and A. Galstyan. A survey on bias and fairness in machine learning. ACM computing surveys (CSUR), 54(6):1–35, 2021. 5

    2021

  28. [36]

    Q. Meng, X. Xu, X. Wang, Y . Qian, Y . Qin, Z. Wang, C. Zhao, F. Zhou, and Z. Lei. Poseface: Pose-invariant features and pose-adaptive loss for face recognition. arXiv preprint arXiv:2107.11721 , 2021. 1

    2021 arXiv

  29. [37]

    Moschoglou, A

    S. Moschoglou, A. Papaioannou, C. Sagonas, J. Deng, I. Kotsia, and S. Zafeiriou. Agedb: the first manually collected, in-the-wild age database. In Proc. CVPR, 2017. 3

    2017

  30. [38]

    Z. Ou, L. Yao, T. Wu, and F. Liu. Aerialface: A light weight framework for unmanned aerial vehicle face recognition. In Proc. FG, 2024. 1

    2024

  31. [39]

    Ozturk, M

    E. Ozturk, M. Prabhushankar, and G. AlRegib. Intelligent multi-view test time augmentation. In Intl. Conf. Image Proc. , 2024. 1

    2024

  32. [40]

    Y . Qian, W. Deng, and J. Hu. Unsupervised face normalization with extreme pose and expression in the wild. In Proc. CVPR, 2019. 1

    2019

  33. [41]

    N. Ruiz, E. Chong, and J. M. Rehg. Fine-grained head pose estimation without keypoints. In Proc. CVPRW, 2018. 2

    2018

  34. [42]

    Schroff, D

    F. Schroff, D. Kalenichenko, and J. Philbin. Facenet: A unified embedding for face recognition and clustering. In Proc. CVPR, 2015. 1

    2015

  35. [43]

    Sengupta, J.-C

    S. Sengupta, J.-C. Chen, C. Castillo, V . M. Patel, R. Chellappa, and D. W. Jacobs. Frontal to profile face verification in the wild. In Proc. WACV, 2016. 3

    2016

  36. [44]

    Shanmugam, D

    D. Shanmugam, D. Blalock, G. Balakrishnan, and J. Guttag. When and why test-time augmentation works. arXiv preprint arXiv:2011.11156 , 1(3):4, 2020. 1

    2011 arXiv

  37. [45]

    Shanmugam, D

    D. Shanmugam, D. Blalock, G. Balakrishnan, and J. Guttag. Better aggregation in test-time augmentation. In Proc. CVPR, 2021. 1, 3

    2021

  38. [46]

    Siarohin, S

    A. Siarohin, S. Lathuili `ere, S. Tulyakov, E. Ricci, and N. Sebe. First order motion model for image animation. In Proc. NeurIPS, 2019. 1

    2019

  39. [47]

    Siarohin, O

    A. Siarohin, O. J. Woodford, J. Ren, M. Chai, and S. Tulyakov. Motion representations for articulated animation. In Proc. NeurIPS, 2021. 1

    2021

  40. [48]

    Sixta, J

    T. Sixta, J. C. Jacques Junior, P. Buch-Cardona, E. Vazquez, and S. Escalera. Fairface challenge at eccv 2020: Analyzing bias in face recognition. In Computer Vision–ECCV 2020 Workshops: Glasgow, UK, August 23–28, 2020, Proceedings, Part VI 16 , pages 463–481. Springer, 2020. 5

    2020

  41. [49]

    Song and F

    Y . Song and F. Wang. Qgface: Quality-guided joint training for mixed- quality face recognition. In Proc. FG, 2024. 1

    2024

  42. [50]

    Y . Sun, C. Cheng, Y . Zhang, C. Zhang, L. Zheng, Z. Wang, and Y . Wei. Circle loss: A unified perspective of pair similarity optimization. In Proc. CVPR, 2020. 1

    2020

  43. [51]

    K. Tian, Y . Jiang, Z. Yuan, B. Peng, and L. Wang. Visual autoregres- sive modeling: Scalable image generation via next-scale prediction. Advances in neural information processing systems , 37:84839–84865,

  44. [52]

    Tolosana, R

    R. Tolosana, R. Vera-Rodriguez, J. Fierrez, A. Morales, and J. Ortega- Garcia. Deepfakes and beyond: A survey of face manipulation and fake detection. Information Fusion , 64:131–148, 2020. 5

    2020

  45. [53]

    L. Tran, X. Yin, and X. Liu. Disentangled representation learning gan for pose-invariant face recognition. In Proc. CVPR, 2017. 1

    2017

  46. [54]

    Van Noorden

    R. Van Noorden. The ethical questions that haunt facial-recognition research. Nature, 587(7834):354–359, 2020. 5

    2020

  47. [55]

    C. Wang, S. Chen, Y . Wu, Z. Zhang, L. Zhou, S. Liu, Z. Chen, Y . Liu, H. Wang, J. Li, et al. Neural codec language models are zero-shot text to speech synthesizers. arXiv preprint arXiv:2301.02111 , 2023. 5

    2023 arXiv

  48. [56]

    F. Wang, J. Cheng, W. Liu, and H. Liu. Additive margin softmax for face verification. Signal Processing Letters , 25(7):926–930, 2018. 1

    2018

  49. [57]

    H. Wang, Y . Wang, Z. Zhou, X. Ji, D. Gong, J. Zhou, Z. Li, and W. Liu. Cosface: Large margin cosine loss for deep face recognition. In Proc. CVPR, 2018. 1

    2018

  50. [58]

    Y . Wang, D. Yang, F. Bremond, and A. Dantcheva. Latent image animator: Learning to animate images via latent space navigation. In Proc. ICLR, 2022. 1

    2022

  51. [59]

    H. Wei, Z. Yang, and Z. Wang. Aniportrait: Audio-driven synthesis of photorealistic portrait animation. arXiv preprint arXiv:2403.17694 ,

    arXiv

  52. [60]

    Y . Xie, H. Xu, G. Song, C. Wang, Y . Shi, and L. Luo. X-portrait: Expressive portrait animation with hierarchical motion attention. In Proc. ACM SIGGRAPH , 2024. 1

    2024

  53. [61]

    D. Yi, Z. Lei, S. Liao, and S. Z. Li. Learning face representation from scratch. arXiv preprint arXiv:1411.7923 , 2014. 1, 3, 5

    2014 arXiv

  54. [62]

    Y . Yin, S. Jiang, J. P. Robinson, and Y . Fu. Dual-attention gan for large-pose face frontalization. In Proc. FG, 2020. 1

    2020

  55. [63]

    B. Zeng, X. Liu, S. Gao, B. Liu, H. Li, J. Liu, and B. Zhang. Face animation with an attribute-guided diffusion model. In Proc. CVPR ,

  56. [64]

    Zheng and W

    T. Zheng and W. Deng. Cross-pose lfw: A database for studying cross-pose face recognition in unconstrained environments. Beijing University of Posts and Telecommunications, Tech. Rep , 5(7):5, 2018. 3

    2018

  57. [65]

    Zheng, W

    T. Zheng, W. Deng, and J. Hu. Cross-age lfw: A database for studying cross-age face recognition in unconstrained environments. arXiv preprint arXiv:1708.08197 , 2017. 3

    2017 arXiv

  58. [66]

    Z. Zhu, G. Huang, J. Deng, Y . Ye, J. Huang, X. Chen, J. Zhu, T. Yang, J. Lu, D. Du, et al. Webface260m: A benchmark unveiling the power of million-scale deep face recognition. In Proc. CVPR, 2021. 1, 3, 5

    2021

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.