Pith. sign in

REVIEW 5 major objections 7 minor 63 references

Towards Large-Scale Pose-Invariant Face Recognition Using Face Defrontalization

T0 review · 5 major / 7 minor · reviewed 2026-08-07 · deepseek-v4-flash

Pith's one-line read By training on synthetic profile faces generated from frontal ones, this paper improves pose-invariant face verification without slowing down inference.

desk verdict Genuinely new training-time defrontalization idea, but the reported effect sizes are within sampling noise and need significance testing. read the letter →

arxiv 2506.04496 v1 pith:JZ5QBLJY submitted 2025-06-04 cs.CV

classification cs.CV
keywords facerecognitionpose-invariantdefrontalizationfrontalizationdataaugmentationArcopticalflowsynthesisalignment
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 aims to improve face recognition under extreme head poses without adding any cost at inference time. Its proposal is face defrontalization: a generative model that turns aligned frontal faces into synthetic 90-degree profile views, used only to augment the training set of a standard ArcFace ResNet-50 feature extractor. The authors argue that defrontalization is an easier task than the usual frontalization because frontal images carry more identity information, and that training on these synthetic profiles teaches pose-invariant features. They report gains on frontal-to-profile benchmarks such as CFP-FP and CPLFW, and larger improvements over their own baseline at Multi-PIE pose angles of 60, 75, and 90 degrees, while keeping the inference pipeline identical. They also argue that the stronger results of some frontalization methods at extreme Multi-PIE poses reflect overfitting of those methods to that small dataset.

What carries the argument

The load-bearing object is the face defrontalization model, an adaptation of the Flow-based Feature Warping Model (FFWM) that takes a 112×112 horizontally bisected aligned frontal face and outputs an aligned 90-degree profile face of the same size. The adaptation removes one encoder and one decoder layer from the forward and backward optical flow networks (cutting them from 52M to 7M parameters) and adds multi-scale mask prediction to control the synthesized background. Training combines pixel, VGG perceptual, adversarial, illumination-preserving, identity-preserving (LightCNN-29), and mask losses. In the feature-extraction stage, defrontalization is applied randomly only to images whose ArcFace alignment error falls below a threshold (about 20 percent of the training set), so the augmenter receives faces it can handle.

What would settle it

Train the same ArcFace ResNet-50 with the same augmentation schedule but replace the learned defrontalization outputs with horizontally mirrored frontal images; if verification on CFP-FP and Multi-PIE ±90 degrees does not drop below the defrontalization-augmented model, the specific benefits of learned synthesis would be falsified. A complementary check is to measure identity preservation of the generated profiles with the baseline feature extractor on the private pairs dataset: if synthetic profiles frequently match the wrong identity, the augmentation is teaching noise.

Watch

Extended reading notes

Core claim

The central claim is that face defrontalization — the inverse of face frontalization — is a valid training-time augmentation for pose-invariant face recognition. Concretely, the paper trains an adapted Flow-based Feature Warping Model (FFWM) on a private frontal-profile pairs dataset of about 50,000 identities, using a new alignment procedure for extreme profile targets that references the corresponding frontal image; then it trains a ResNet-50 with ArcFace loss on MS1MV2 while randomly defrontalizing roughly 20 percent of frontal images (choosing left or right side) to 90-degree profile views. The resulting feature extractor improves verification accuracy over the no-augmentation baseline on AgeDB-30, CFP-FP, CFP-FF, CALFW, and CPLFW, and lifts Multi-PIE top-1 accuracy at ±75 degrees from 95.64 to 97 percent and at ±90 degrees from 70.03 to 78.6 percent, with no change to the deployment pipeline and the same 5 ms inference speed.

Load-bearing premise

The private frontal-profile pairs dataset, aligned with the new frontal-referenced procedure, is good enough that the reduced-capacity defrontalization generator (optical flow cut from 52M to 7M parameters) produces identity-preserving 90-degree profiles; if the synthetic profiles leak artifacts or lose identity, the augmentation could teach dataset-specific distortions instead of pose invariance.

Editorial extensions

If this is right

  • Training-time defrontalization yields pose-invariant features while the deployed pipeline stays a single ResNet-50 at 5 ms per image, versus 11 ms for frontalization-plus-encoding.
  • The gains transfer from the trained 90-degree pose to intermediate poses, since Multi-PIE accuracy at ±60 and ±75 improves even though no 60- or 75-degree images were synthesized.
  • Cross-pose benchmarks improve most where pose difference is largest, with CFP-FP (mean yaw difference 68.67 degrees) rising from 97.9 to 98.114 percent and CPLFW from 92.66 to 92.967 percent.
  • The poor transfer of the FFWM frontalization model to CFP-FP (85.15 percent versus 98.114 percent for the defrontalized model) supports the paper's claim that some frontalization methods are overfitted to Multi-PIE.
  • A publicly released trained defrontalization model lets others reproduce the augmentation without access to the private pairs dataset.

Reading between the lines

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

  • The same trick should extend to synthesizing multiple poses (e.g., 45, 60, 75 degrees) from one frontal image, with the augmentation schedule choosing a random target angle; the paper's gains at untrained intermediate poses suggest this would further smooth pose response.
  • Defrontalization could be combined with frontalization of genuinely non-frontal training samples, making the training distribution bilaterally pose-normalized rather than just enriched; this is a testable extension the paper does not explore.
  • Because the augmentation is decoupled from the feature extractor, better identity-preserving generators (including diffusion-based ones) could be dropped into the pipeline and should translate directly into larger pose-invariance gains at no inference cost.
  • The overfitting argument implies a concrete test: retrain frontalization models on a large multi-pose dataset and compare; if they then beat defrontalization on extreme poses, the remaining advantage of defrontalization would be purely its zero inference overhead.
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

5 major / 7 minor

Summary. The paper proposes face defrontalization as a training-time augmentation for pose-invariant face recognition. A modified FFWM model is trained on a private frontal-profile pairs dataset, using a newly proposed landmark-based alignment procedure for profile images. The trained model is then used to randomly defrontalize frontal training images from MS1MV2, and a ResNet-50 with ArcFace loss is trained on the augmented data. At inference, no generative model is used. Experiments on LFW, AgeDB, CFP, CALFW, CPLFW, and Multi-PIE compare the augmented model against a baseline without augmentation and against test-time FFWM frontalization. The results show small gains on most datasets and larger gains on Multi-PIE extreme poses, with no inference overhead.

Significance. If the reported gains are real, the defrontalization augmentation offers a simple, practical way to improve pose robustness without changing the inference pipeline, and the paper's release of the trained defrontalization model is a useful resource. The idea of learning the inverse of frontalization is a sensible alternative to test-time frontalization. However, the evidence is currently not persuasive: the accuracy differences are small relative to sampling noise, no significance testing is provided, the comparison with frontalization is uncontrolled, and the private training data makes the core generative component unverifiable. With additional statistical rigor and controlled comparisons, the contribution could become a useful data-augmentation technique for the community.

major comments (5)
  1. [§4.2, Tables 2 and 3] The central empirical claim is not supported by significance testing. The paper reports single runs without error bars or repeated seeds. For the in-the-wild benchmarks, the test set sizes are 6,000–7,000 pairs; at accuracy around 98%, the standard error is approximately 0.17–0.18 percentage points, so the largest reported gains (CFP-FP +0.214 pp, CPLFW +0.307 pp) are within about two standard errors, and LFW shows a small negative change. On Multi-PIE Setting 2 (n=137), the ±75° gain of 1.36 pp corresponds to roughly two correct answers, and the ±90° gain of 8.57 pp has a confidence interval around the baseline of roughly ±7.7 pp. The authors should report confidence intervals, paired significance tests (e.g., McNemar), and ideally multiple training runs, before claiming improvement.
  2. [§4.2, Table 2] The comparison with FFWM frontalization is not controlled. The rows 'FFWM, ResNet-50 Baseline' and 'FFWM, ResNet-50 & defrontalization' apply FFWM frontalization only at test time, while the feature extractor is trained without any frontalized or defrontalized data. The defrontalization method, by contrast, changes the training distribution. Thus the accuracy gap between these rows could be due to training-time vs test-time use of a generative model rather than to the choice of defrontalization over frontalization. A controlled baseline that trains with FFWM frontalized data as augmentation is needed to support the claim of 'clear superiority' over FFWM.
  3. [§3.1, §3.2, §4.2] The two methodological contributions—the alignment method and the defrontalization model itself—lack quantitative evaluation. Section 3.1 states the alignment was validated only 'through visualization,' and Figure 5 provides only qualitative examples. Because the training pairs are private (Section 3.4), readers cannot assess identity preservation or consistency of the generated 90° profiles. The paper should provide quantitative measures: landmark alignment error on a public profile dataset (e.g., using the proposed procedure), and an identity-preservation metric (e.g., verification accuracy or cosine similarity on synthetic profiles) for the defrontalization model. Releasing the alignment code and a small de-identified sample of pairs would also help.
  4. [§4.2, Table 3] The Multi-PIE comparison with related work mixes backbones, training data, and protocols. The authors use ResNet-50 trained on MS1MV2, while most cited methods (e.g., LightCNN, DR-GAN, FF-GAN) use different architectures and training sets. The claim that only a few frontalization methods outperform at extreme poses is therefore not a meaningful comparison. The authors should restrict the comparison to same-backbone baselines or clearly state that cross-method numbers are not directly comparable.
  5. [§5 and Abstract] The inference that 'at least some of the current methods may be overfitted to small datasets' is not directly supported. The experiments do not manipulate training-set size or show a correlation between performance gap and dataset size; they only show that the proposed model does not use Multi-PIE training identities. This claim should be softened or backed by an additional experiment, such as training the same backbone with and without Multi-PIE training data.
minor comments (7)
  1. [§3.3] The alignment error threshold is not specified; the reader only knows that roughly 20% of images were defrontalized. Please report the actual threshold or the distribution of alignment errors.
  2. [§3.2, Eq. (7)] The λ weights are said to follow the original FFWM setting, but the values are not listed. Provide them for reproducibility, along with the value of λ5 (stated as 1).
  3. [§4.1] The GPU is referred to as 'NVIDIA GeForce GTX 2070 Super'; the correct product name is 'RTX 2070 Super'.
  4. [References] Reference [5] contains 'B. F. Klare ajd B. Klein'; 'ajd' should be 'and'.
  5. [§4.2, Table 3] The table would benefit from indicating which methods use the same Multi-PIE Setting 2 protocol; the caption currently relies on the text.
  6. [§3.1] The description of the third alignment step is somewhat confusing: 'we set the y-coordinate of this landmark to the transformed value from the previous step'—please clarify whether this is the y-coordinate of the ear landmark after the transformation in step 2.
  7. [§5, Table 4] The inference speed numbers in Table 4 are reported without specifying hardware; please state the platform, since times are only meaningful relative to it.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the defrontalization model, augmentation training, and held-out benchmarks form an independent chain; no self-citation is load-bearing.

full rationale

The paper's claimed derivation chain is self-contained and not circular. The defrontalization FFWM model is an independent component trained on a private frontal–profile pairs dataset (Section 3.2), and the ResNet-50 feature extractor is trained on MS1MV2 with random defrontalization augmentation (Section 3.3); neither training process uses the LFW, AgeDB, CFP, or Multi-PIE test pairs, so the evaluations in Tables 2–3 are held-out external benchmarks. No fitted parameter is renamed as a prediction: the alignment-error threshold (~20% augmentation) is chosen on the training set, and the reported accuracies are measured on external test data. There are no self-citations at all: the method relies on external prior work (FFWM [45], ArcFace [10], InsightFace [2]), so no uniqueness theorem or ansatz is imported from the authors' own prior papers. The Identity Preserving Loss (Eq. 5) uses a pretrained LightCNN-29 as a training regularizer, but the central claim — that augmentation improves held-out pose-invariant verification — does not reduce to that loss. The paper even reports its own failure mode (no clear superiority on Multi-PIE at ±75/±90, stated in the abstract and visible in Table 3), which is inconsistent with result-forcing. The skeptic's concerns about small gains without error bars (Section 4.2) and the visualization-only validation of the alignment procedure (Section 3.1) are correctness and robustness risks, not circularity. The only structural closeness is that the same FFWM architecture family generates both the augmentation and the frontalization baseline, but this is an experimental setup choice, not a definitional equivalence. Score 0 reflects the absence of any circular step per the quoted-equation standard.

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

The method relies on standard deep learning assumptions (landmark accuracy, label reliability) plus a novel alignment procedure validated visually. No new physical entities are introduced. The main burden is the private dataset and the transferred FFWM design.

free parameters (4)
  • Alignment error threshold for defrontalization = Set so about 20% of MS1MV2 images are defrontalized
    Controls how many training images receive synthetic profile augmentation; no ablation or principled selection is reported.
  • Defrontalization training epochs = 50
    Reduced from FFWM's 200 because private dataset has more identities; convergence criterion is not formally justified.
  • Mask loss weight lambda_5 = 1
    Added for the new multi-scale mask prediction, assigned unit weight without ablation.
  • Defrontalization side selection = Left or right randomly
    Augmentation randomly defrontalizes left or right half of the face; no analysis of effect on accuracy.
assumptions (4)
  • domain assumption FFWM's architecture and loss decomposition transfer to the inverse defrontalization task.
    Section 3.2 adapts FFWM optical flow, generator, discriminator, and IPM, and reuses loss weights from the original paper without revalidation for the inverse direction.
  • domain assumption Pretrained LightCNN-29 provides a reliable identity-preserving signal for synthetic profile faces.
    Identity preserving loss in Eq. 5 uses LightCNN-29 features; if the pretrained model is weak on profiles, the generator may not preserve identity.
  • ad hoc to paper Proposed face alignment using frontal-reference landmarks yields consistent profile alignment across the private dataset.
    Section 3.1 introduces a novel alignment validated 'through visualization' only, not with a quantitative landmark consistency metric.
  • domain assumption Face++ landmark detection is accurate enough on frontal and profile images.
    Alignment relies on Face++ dense landmarks; no accuracy analysis on extreme profiles is given.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Towards Large-Scale Pose-Invariant Face Recognition Using Face Defrontalization." pith.science (2026). https://pith.science/paper/JZ5QBLJY

@misc{pith2026250604496,
  author       = {Pith},
  title        = {Pith review of: Towards Large-Scale Pose-Invariant Face Recognition Using Face Defrontalization},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JZ5QBLJY}},
  note         = {Machine review of arXiv:2506.04496}
}
read the original abstract

Face recognition under extreme head poses is a challenging task. Ideally, a face recognition system should perform well across different head poses, which is known as pose-invariant face recognition. To achieve pose invariance, current approaches rely on sophisticated methods, such as face frontalization and various facial feature extraction model architectures. However, these methods are somewhat impractical in real-life settings and are typically evaluated on small scientific datasets, such as Multi-PIE. In this work, we propose the inverse method of face frontalization, called face defrontalization, to augment the training dataset of facial feature extraction model. The method does not introduce any time overhead during the inference step. The method is composed of: 1) training an adapted face defrontalization FFWM model on a frontal-profile pairs dataset, which has been preprocessed using our proposed face alignment method; 2) training a ResNet-50 facial feature extraction model based on ArcFace loss on a raw and randomly defrontalized large-scale dataset, where defrontalization was performed with our previously trained face defrontalization model. Our method was compared with the existing approaches on four open-access datasets: LFW, AgeDB, CFP, and Multi-PIE. Defrontalization shows improved results compared to models without defrontalization, while the proposed adjustments show clear superiority over the state-of-the-art face frontalization FFWM method on three larger open-access datasets, but not on the small Multi-PIE dataset for extreme poses (75 and 90 degrees). The results suggest that at least some of the current methods may be overfitted to small datasets.

Figures

Figures reproduced from arXiv: 2506.04496 by the authors.

Figure 1
Figure 1. The main components of the face feature ex [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Proposed methodology for a large-scale pose-invariant face recognition [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Visualization of our proposed face alignment [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Input and outputs visualization of our face de [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Qualitative results of the proposed face de [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

63 extracted references · 62 canonical work pages

  1. [1]

    Face++.https://www.faceplusplus.com/. 4

  2. [2]

    Insightface.https://github.com/ deepinsight/insightface. 4, 7

  3. [3]

    Face recognition using deep 10 multi-pose representations.2016 IEEE Winter Conference on Applications of Computer Vision (WACV), 2016

    Wael AbdAlmageeda, Yue Wu, Stephen Rawls, Shai Harel, Tal Hassner, Iacopo Masi, Jongmoo Choi, Jatu- porn Toy Leksut, Jungyeon Kim, Prem Natarajana, Ram Nevatia, and Gerard Medioni. Face recognition using deep 10 multi-pose representations.2016 IEEE Winter Conference on Applications of Computer Vision (WACV), 2016. 3

  4. [4]

    Past, present, and future of face recognition: A review.Electronics, 9(8), 2020

    Insaf Adjabi, Abdeldjalil Ouahabi, Amir Benzaoui, and Abdelmalik Taleb-Ahmed. Past, present, and future of face recognition: A review.Electronics, 9(8), 2020. 1

  5. [5]

    B. F. Klare ajd B. Klein, E. Taborsky, A. Blanton, J. Ch- eney, K. Allen, P. Grother, A. Mah, and A. K. Jain. Iarpa janus benchmark-b face dataset.2017 IEEE Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), 2017. 2

  6. [6]

    Learning a high fidelity pose invariant model for high-resolution face frontalization.Advances in Neural In- formation Processing Systems, page 2867–2877, 2018

    Jie Cao, Yibo Hu, Hongwen Zhang, Ran He, and Zhenan Sun. Learning a high fidelity pose invariant model for high-resolution face frontalization.Advances in Neural In- formation Processing Systems, page 2867–2877, 2018. 3, 10

  7. [7]

    Pose-robust face recognition via deep residual equivariant mapping.Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 5187–5196, 2018

    Kaidi Cao, Yu Rong, Cheng Li, Xiaoou Tang, and Chen Change Loy. Pose-robust face recognition via deep residual equivariant mapping.Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 5187–5196, 2018. 3

  8. [8]

    Parkhi, and Andrew Zisserman

    Qiong Cao, Li Shen, Weidi Xie, Omkar M. Parkhi, and Andrew Zisserman. Vggface2: A dataset for recognis- ing faces across pose and age.2018 13th IEEE Interna- tional Conference on Automatic Face & Gesture Recogni- tion (FG 2018), 2018. 4

Show all 63 references
  1. [9]

    Uv-gan: Adversarial fa- cial uv map completion for pose-invariant face recognition

    Jiankang Deng, Shiyang Cheng, Niannan Xue, Yuxiang Zhou, and Stefanos Zafeiriou. Uv-gan: Adversarial fa- cial uv map completion for pose-invariant face recognition. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 7093–7102, 2018. 3

  2. [10]

    Arcface: Additive angular margin loss for deep face recognition.Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 4690–4699, 2019

    Jiankang Deng, Jia Guo, Niannan Xue, and Stefanos Zafeiriou. Arcface: Additive angular margin loss for deep face recognition.Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 4690–4699, 2019. 2, 3, 4, 7

  3. [11]

    Marginal loss for deep face recognition.Proceedings of the IEEE Conference on Computer Vision and Pattern Recog- nition (CVPR) Workshops, pages 60–68, 2017

    Jiankang Deng, Yuxiang Zhou, and Stefanos Zafeiriou. Marginal loss for deep face recognition.Proceedings of the IEEE Conference on Computer Vision and Pattern Recog- nition (CVPR) Workshops, pages 60–68, 2017. 4

  4. [12]

    Generative adversarial networks

    Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Generative adversarial networks. Communications of the ACM, 63:139–144, 2020. 3

  5. [13]

    Gross, I

    R. Gross, I. Matthews, J. Cohn, T. Kanade, and S. Baker. Multi-pie.Image and Vision Computing, 28:807–813,

  6. [14]

    Ms-celeb-1m: A dataset and benchmark for large-scale face recognition.European Conference on Computer Vision (ECCV), page 87–102, 2016

    Yandong Guo, Lei Zhang, Yuxiao Hu, Xiaodong He, and Jianfeng Gao. Ms-celeb-1m: A dataset and benchmark for large-scale face recognition.European Conference on Computer Vision (ECCV), page 87–102, 2016. 4, 7

  7. [15]

    Ef- fective face frontalization in unconstrained images.Pro- ceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 4295–4304, 2015

    Tal Hassner, Shai Harel, Eran Paz, and Roee Enbar. Ef- fective face frontalization in unconstrained images.Pro- ceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 4295–4304, 2015. 3

  8. [16]

    Deep residual learning for image recognition.Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 770–778, 2016

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition.Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 770–778, 2016. 7

  9. [17]

    Enhancing face recognition with self-supervised 3d recon- struction.Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 4062–4071, 2022

    Mingjie He, Jie Zhang, Shiguang Shan, and Xilin Chen. Enhancing face recognition with self-supervised 3d recon- struction.Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 4062–4071, 2022. 3

  10. [18]

    Pose-guided photorealistic face rotation.15th IEEE In- ternational Conference on Automatic Face and Gesture Recognition, page 8398–8406, 2018

    Yibo Hu, Xiang Wu, Bing Yu, Ran He, and Zhenan Sun. Pose-guided photorealistic face rotation.15th IEEE In- ternational Conference on Automatic Face and Gesture Recognition, page 8398–8406, 2018. 3, 10

  11. [19]

    Labeled faces in the wild: A database forstudying face recognition in unconstrained environ- ments.ECCV Workshop on Faces in Real-life Images,

    Gary B Huang, Marwan Mattar, Tamara Berg, and Eric Learned-Miller. Labeled faces in the wild: A database forstudying face recognition in unconstrained environ- ments.ECCV Workshop on Faces in Real-life Images,

  12. [20]

    Rui Huang, Shu Zhang, Tianyu Li, and Ran He. Beyond face rotation: Global and local perception gan for photore- alistic and identity preserving frontal view synthesis.Pro- ceedings of the IEEE International Conference on Com- puter Vision, page 2439–2448, 2017. 2, 3, 10

  13. [21]

    Yu-Hsuan Huang and Homer H. Chen. Face recognition under low illumination via deep feature reconstruction net- work. In2020 IEEE International Conference on Image Processing (ICIP), pages 2161–2165, 2020. 1

  14. [22]

    Yeong-Joon Ju, Gun-Hee Lee, Jung-Ho Hong, and Seong- Whan Lee. Complete face recovery gan: Unsupervised joint face rotation and de-occlusion from a single-view image.Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), pages 3711– 3721, 2022. 3

  15. [23]

    Meina Kan, Shiguang Shan, Hong Chang, and Xilin Chen. Stacked progressive auto-encoders (spae) for face recogni- tion across poses.Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 1883–1890, 2014. 2, 3

  16. [24]

    Face recognition systems: A survey.Sensors, 20(2):342, 2020

    Yassin Kortli, Maher Jridi, Ayman Al Falou, and Mo- hamed Atri. Face recognition systems: A survey.Sensors, 20(2):342, 2020. 1

  17. [25]

    Li and Anil K

    Stan Z. Li and Anil K. Jain. Handbook of face recognition. Springer, 2011. 1, 2 11

  18. [26]

    Sphereface: Deep hypersphere em- bedding for face recognition.Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 212–220, 2017

    Weiyang Liu, Yandong Wen, Zhiding Yu, Ming Li, Bhik- sha Raj, and Le Song. Sphereface: Deep hypersphere em- bedding for face recognition.Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 212–220, 2017. 4

  19. [27]

    Learning pose-aware models for pose- invariant face recognition in the wild.IEEE Transactions on Pattern Analysis and Machine Intelligence, 41:379– 393, 2018

    Iacopo Masi, Feng-Ju Chang, Jongmoo Choi, Shai Harel, Jungyeon Kim, KangGeon Kim, Jatuporn Leksut, Stephen Rawls, Yue Wu, Tal Hassner, Wael AbdAlmageed, Ger- ard Medioni, Louis-Philippe Morency, Prem Natarajan, and Ram Nevatia. Learning pose-aware models for pose- invariant fa...

  20. [28]

    Pose-aware face recognition in the wild.Pro- ceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 4838–4846, 2016

    Iacopo Masi, Stephen Rawls, Gerard Medioni, and Prem Natarajan. Pose-aware face recognition in the wild.Pro- ceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 4838–4846, 2016. 2, 3

  21. [29]

    Agedb: The first manually collected, in-the- wild age database.Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Work- shops, pages 51–59, 2017

    Stylianos Moschoglou, Athanasios Papaioannou, Chris- tos Sagonas, Jiankang Deng, Irene Kotsia, and Stefanos Zafeiriou. Agedb: The first manually collected, in-the- wild age database.Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Work- shop...

  22. [30]

    Parkhi, Andrea Vedaldi, and Andrew Zisser- man

    Omkar M. Parkhi, Andrea Vedaldi, and Andrew Zisser- man. Deep face recognition.BMVC 2015 - Proceedings of the British Machine Vision Conference, 2015. 4

  23. [31]

    Unsupervised face normalization with extreme pose and expression in the wild.Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition, pages 9851–9858,

    Yichen Qian, Weihong Deng, and Jiani Hu. Unsupervised face normalization with extreme pose and expression in the wild.Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition, pages 9851–9858,

  24. [32]

    Synface: Face recognition with synthetic data.Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 10880– 10890, 2021

    Haibo Qiu, Baosheng Yu, Dihong Gong, Zhifeng Li, Wei Liu, and Dacheng Tao. Synface: Face recognition with synthetic data.Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 10880– 10890, 2021. 2

  25. [33]

    Robust statistical face frontalization

    Christos Sagonas, Yannis Panagakis, Stefanos Zafeiriou, and Maja Pantic. Robust statistical face frontalization. Proceedings of the IEEE International Conference on Computer Vision, pages 3871–3879, 2015. 3

  26. [34]

    Della Santina, Valeria Potyagaylo, Americo A

    Charles C. Della Santina, Valeria Potyagaylo, Americo A. Migliaccio, Lloyd B. Minor, and John P. Carey. Orien- tation of human semicircular canals measured by three- dimensional multiplanar ct reconstruction.Journal of the Association for Research in Otolaryngology, 6:191–206,

  27. [35]

    Facenet: A unified embedding for face recognition and clustering.Proceedings of the IEEE Conference on Com- puter Vision and Pattern Recognition (CVPR), pages 815– 823, 2015

    Florian Schroff, Dmitry Kalenichenko, and James Philbin. Facenet: A unified embedding for face recognition and clustering.Proceedings of the IEEE Conference on Com- puter Vision and Pattern Recognition (CVPR), pages 815– 823, 2015. 4

  28. [36]

    Patel, Rama Chellappa, and David W

    Soumyadip Sengupta, Jun-Cheng Chen, Carlos Castillo, Vishal M. Patel, Rama Chellappa, and David W. Jacobs. Frontal to profile face verification in the wild.Winter Conference on Applications of Computer Vision (WACV),

  29. [37]

    Very deep convolutional networks for large-scale image recognition

    Karen Simonyan and Andrew Zisserman. Very deep convolutional networks for large-scale image recognition. arXiv preprint arXiv:1409.1556, 2014. 6

  30. [38]

    Yaniv Taigman, Ming Yang Marc’Aurelio Ranzato, and Lior Wolf. Deepface: Closing the gap to human-level per- formance in face verification.Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 1701–1708, 2014. 4

  31. [39]

    Disentangled repre- sentation learning gan for pose-invariant face recognition

    Luan Tran, Xi Yin, and Xiaoming Liu. Disentangled repre- sentation learning gan for pose-invariant face recognition. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 1415–1424, 2017. 10

  32. [40]

    Additive margin softmax for face verification.IEEE Signal Processing Letters, 25:926–930, 2018

    Feng Wang, Weiyang Liu, Haijun Liu, and Jian Cheng. Additive margin softmax for face verification.IEEE Signal Processing Letters, 25:926–930, 2018. 4

  33. [41]

    Pseudo facial generation with extreme poses for face recognition.Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 1994–2003, 2021

    Guoli Wang, Jiaqi Ma, Qian Zhang, Jiwen Lu, and Jie Zhou. Pseudo facial generation with extreme poses for face recognition.Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 1994–2003, 2021. 3

  34. [42]

    Cosface: Large margin cosine loss for deep face recognition.Pro- ceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 5265–5274, 2018

    Hao Wang, Yitong Wang, Zheng Zhou, Xing Ji, Dihong Gong, Jingchao Zhou, Zhifeng Li, and Wei Liu. Cosface: Large margin cosine loss for deep face recognition.Pro- ceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 5265–5274, 2018. 4

  35. [43]

    Deep face recognition: A survey.Neurocomputing, Elsevier, 429:215–244, 2021

    Mei Wang and Weihong Deng. Deep face recognition: A survey.Neurocomputing, Elsevier, 429:215–244, 2021. 2

  36. [44]

    Anti-occlusion face recognition algorithm based on a deep convolutional neural network

    Xi Wang and Wei Zhang. Anti-occlusion face recognition algorithm based on a deep convolutional neural network. Computers & Electrical Engineering, 96:107461, 2021. 1

  37. [45]

    Learning flow-based feature warping for face frontalization with illumination inconsistent supervision.European Conference on Com- puter Vision 2020, page 558–574, 2020

    Yuxiang Wei, Ming Liu, Haolin Wang, Ruifeng Zhu, Gu- osheng Hu, and Wangmeng Zuo. Learning flow-based feature warping for face frontalization with illumination inconsistent supervision.European Conference on Com- puter Vision 2020, page 558–574, 2020. 2, 3, 4, 5, 8, 10

  38. [46]

    A discriminative feature learning approach for deep face recognition.European Conference on Computer Vision (ECCV) 2016, page 499–515, 2016

    Yandong Wen, Kaipeng Zhang, Zhifeng Li, and Yu Qiao. A discriminative feature learning approach for deep face recognition.European Conference on Computer Vision (ECCV) 2016, page 499–515, 2016. 4

  39. [47]

    Cashman, and Jamie Shotton

    Erroll Wood, Tadas Baltru ˇsaitis, Charlie Hewitt, Sebastian Dziadzio, Thomas J. Cashman, and Jamie Shotton. Fake 12 it till you make it: Face analysis in the wild using syn- thetic data alone.Proceedings of the IEEE/CVF Inter- national Conference on Computer Vision (ICCV), pa...

  40. [48]

    A light cnn for deep face representation with noisy labels

    Xiang Wu, Ran He, Zhenan Sun, and Tieniu Tan. A light cnn for deep face representation with noisy labels. IEEE Transactions on Information Forensics and Security, 13:2884–2896, 2018. 7, 10

  41. [49]

    Condi- tional convolutional neural network for modality-aware face recognition.Proceedings of the IEEE International Conference on Computer Vision (ICCV), pages 3667– 3675, 2015

    Chao Xiong, Xiaowei Zhao, Danhang Tang, Karlekar Jayashree, Shuicheng Yan, and Tae-Kyun Kim. Condi- tional convolutional neural network for modality-aware face recognition.Proceedings of the IEEE International Conference on Computer Vision (ICCV), pages 3667– 3675, 2015. 3

  42. [50]

    Multi-view face synthesis via pro- gressive face flow.IEEE Transactions on Image Process- ing, 30:6024–6035, 2021

    Yangyang Xu, Xuemiao Xu, Jianbo Jiao, Keke Li, Cheng Xu, and Shengfeng He. Multi-view face synthesis via pro- gressive face flow.IEEE Transactions on Image Process- ing, 30:6024–6035, 2021. 3

  43. [51]

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

  44. [52]

    Rotating your face using multi-task deep neural network.Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 676–684, 2015

    Junho Yim, Heechul Jung, ByungIn Yoo, Changkyu Choi, Dusik Park, and Junmo Kim. Rotating your face using multi-task deep neural network.Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 676–684, 2015. 3

  45. [53]

    Multi-task convolutional neural network for pose-invariant face recognition.IEEE Trans- actions on Image Processing, pages 964–975, 2017

    Xi Yin and Xiaoming Liu. Multi-task convolutional neural network for pose-invariant face recognition.IEEE Trans- actions on Image Processing, pages 964–975, 2017. 3

  46. [54]

    Towards large-pose face frontalization in the wild.Proceedings of the IEEE International Confer- ence on Computer Vision (ICCV), pages 3990–3999, 2017

    Xi Yin, Xiang Yu, Kihyuk Sohn, Xiaoming Liu, and Man- mohan Chandraker. Towards large-pose face frontalization in the wild.Proceedings of the IEEE International Confer- ence on Computer Vision (ICCV), pages 3990–3999, 2017. 3, 10

  47. [55]

    Robinson, and Yun Fu

    Yu Yin, Songyao Jiang, Joseph P. Robinson, and Yun Fu. Dual-attention gan for large-pose face frontalization.Pro- ceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2020. 3, 10

  48. [56]

    Range loss for deep face recognition with long-tailed training data.Proceedings of the IEEE Inter- national Conference on Computer Vision (ICCV), pages 5409–5418, 2017

    Xiao Zhang, Zhiyuan Fang, Yandong Wen, Zhifeng Li, and Yu Qiao. Range loss for deep face recognition with long-tailed training data.Proceedings of the IEEE Inter- national Conference on Computer Vision (ICCV), pages 5409–5418, 2017. 4

  49. [57]

    Pose-invariant face recogni- tion via adaptive angular distillation.Proceedings of the AAAI Conference on Artificial Intelligence, 36(3):3390– 3398, Jun

    Zhenduo Zhang, Yongru Chen, Wenming Yang, Guijin Wang, and Qingmin Liao. Pose-invariant face recogni- tion via adaptive angular distillation.Proceedings of the AAAI Conference on Artificial Intelligence, 36(3):3390– 3398, Jun. 2022. 1, 2, 3

  50. [58]

    Towards pose invariant face recognition in the wild.Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 2207–2216, 2018

    Jian Zhao, Yu Cheng, Yan Xu, Lin Xiong, Jianshu Li, Fang Zhao, Karlekar Jayashree, Sugiri Pranata, Shengmei Shen, Junliang Xing, Shuicheng Yan, and Jiashi Feng. Towards pose invariant face recognition in the wild.Proceedings of the IEEE Conference on Computer Vision and Patter...

  51. [59]

    3d-aided deep pose-invariant face recognition.IJCAI, 2:11, 2018

    Jian Zhao, Lin Xiong, Yu Cheng, Yi Cheng, Jianshu Li, Li Zhou, Yan Xu, Jayashree Karlekar, Sugiri Pranata, Sheng- mei Shen, Junliang Xing, Shuicheng Yan, and Jiashi Feng. 3d-aided deep pose-invariant face recognition.IJCAI, 2:11, 2018. 10

  52. [60]

    W. Zhao, R. Chellappa, P. J. Phillips, and A. Rosenfeld. Face recognition: A literature survey.ACM Comput. Surv., 35(4):399–458, 2003. 1

  53. [61]

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

    Tianyue Zheng and Weihong Deng. Cross-pose lfw: A database for studying cross-pose face recognition in un- constrained environments. 2018. 7

  54. [62]

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

    Tianyue Zheng, Weihong Deng, and Jiani Hu. Cross-age lfw: A database for studying cross-age face recognition in unconstrained environments. 2017. 7

  55. [63]

    Xiangyu Zhu, Zhen Lei, Junjie Yan, Dong Yi, and Stan Z. Li. High-fidelity pose and expression normalization for face recognition in the wild.Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 787–796, 2015. 2, 3 13

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.