Pith. sign in

REVIEW 4 major objections 5 minor 38 references

Precise Facial Landmark Detection by Dynamic Semantic Aggregation Transformer

T0 review · 4 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read A dynamic transformer that activates different feature channels per face reports state-of-the-art facial landmark accuracy on four benchmarks.

desk verdict The architecture is a reasonable incremental contribution, but the paper's own ablation contradicts its implementation description, making the main results hard to attribute. read the letter →

arxiv 2412.00740 v1 pith:ET3XT3QE submitted 2024-12-01 cs.CV

Jun Wan , He Liu , Yujia Wu , Zhihui Lai , Wenwen Min , Jun Liu This is my paper · ORCID
classification cs.CV
keywords faciallandmarkdetectionfacealignmentdynamicneuralnetworkscross-channelattentionheatmapregressionsamplepartitioningsemanticgapocclusionrobustness
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

Facial landmark detection degrades on faces with large pose, occlusion, or makeup because conventional models route every image through the same fixed network. This paper introduces a Dynamic Semantic-Aggregation Transformer (DSAT) that decides per image which feature channels to activate, grouping similar faces into subsets that share a computational path, and then uses a cross-channel attention module to fuse multi-scale features so that each subset learns specialized representations. On standard benchmarks the method reports lower normalized mean error than prior state-of-the-art models (2.86 on 300W full set, 4.12 on WFLW, 2.57 on COFW with 0% failure, 1.35 on AFLW) and shows that harder faces trigger more active channels. If the reported gains hold, the approach offers a general recipe for making dense prediction networks adaptive to sample difficulty rather than optimized for the average case.

What carries the argument

Two modules carry the argument. The Dynamic Semantic-Aware (DSA) model computes a per-channel descriptor by adaptive average pooling, adds Gaussian noise, then passes it through a saturating sigmoid and a binary threshold to produce a binary gate per channel; samples whose gates look alike are treated as one subset and share a pathway, and the gating is trained with an Improved SemHash-style straight-through estimator so gradients can flow through the discrete choice. The Dynamic Semantic Specialization (DSS) model wraps an hourglass backbone with a Cross-Channel Attention (CCA) module: multi-scale features are flattened into sequences, and features at each scale issue queries against keys and values built from all scales, so different scales update each other and semantic gaps between them are reduced. Together the modules instantiate dynamic architecture (which channels are used) and dynamic parameters (how multi-scale features are aggregated).

What would settle it

Train DSAT with the channel descriptor d replaced by random noise (or by fixed random gates) and measure NME on 300W and WFLW; if accuracy stays close to the reported DSAT numbers while the gating is meaningless, then the semantic-correlation assumption is not the source of the improvement. Alternatively, compare gate assignments for images of the same face under different poses and check whether they cluster as the paper's subset-partitioning story requires.

Watch

Extended reading notes

Core claim

The paper's central claim is that making the network itself sample-aware—switching on different feature channels for different faces—lets a single model learn features specialized to easy and hard subsets, and that this beats fixed-architecture baselines. Concretely DSAT reports NMEio of 2.86 on 300W Fullset, 4.12 on WFLW, NMEio 2.57 with 0.00 failure rate on COFW, and NMEdiag 1.35 on AFLW, each lower than the listed comparison methods. The authors interpret the improvement as evidence that sample partitioning by channel-level semantic correlation yields more discriminative representations than a single shared pathway.

Load-bearing premise

The method assumes that a single pooled channel vector, after adding Gaussian noise and thresholding, genuinely captures which facial features are important for a given image, so that similar gate patterns correspond to genuinely similar faces; if that mapping is arbitrary, the gating is random and the reported gains would have to come from elsewhere.

Editorial extensions

If this is right

  • If the reported results are correct, a single face alignment network can serve both easy and hard images without a separate expert model per subset.
  • The observed pattern that harder faces activate more channels implies the gate vector carries a usable difficulty signal, so network capacity is spent where the error risk is higher.
  • Cross-channel attention across scales offers a concrete way to reduce the semantic ambiguity that plain skip connections introduce in hourglass-style detectors.
  • The same dynamic gating and cross-scale fusion recipe could transfer to other dense prediction tasks with hard subpopulations, such as pose estimation or medical image segmentation.

Reading between the lines

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

  • The gate vector could be read as an unsupervised difficulty scorecard, enabling uncertainty estimates or active labeling for hard cases; the paper does not test this directly.
  • Because the gating uses a fixed saturating sigmoid and binary rule, the method's gains might depend on the noise injection schedule; replacing the hand-set schedule with a learned one is a natural test of whether the semantic-correlation story is load-bearing.
  • The CCA module's per-scale queries essentially implement a learned version of feature pyramids; applying it to other multi-scale fusion problems, such as semantic segmentation, is a direct transfer the paper leaves implicit.
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

4 major / 5 minor

Summary. The paper proposes DSAT, a dynamic face-alignment architecture that combines a Dynamic Semantic-Aware (DSA) gating module, which partitions samples by estimating channel-level semantic relevance, and a Dynamic Semantic Specialization (DSS) module with cross-channel attention to fuse multi-scale features. The model is evaluated on 300W, WFLW, COFW, and AFLW, reporting state-of-the-art or near-state-of-the-art NME values, and the authors provide qualitative evidence that harder samples activate more channels. The central claim is that this dynamic gating and multi-scale aggregation improves landmark accuracy over fixed-architecture baselines.

Significance. If the reported results are reproducible and attributable to the described architecture, DSAT would be a useful addition to face alignment, particularly for occlusion and pose robustness. The paper has concrete strengths: it releases code, evaluates on four standard benchmarks, and includes ablations and qualitative channel-activation analyses. The proposed mechanisms—sample-dependent channel gating and cross-channel multi-scale attention—are clearly relevant to the field. However, the empirical support is weakened by missing error bars, by several benchmark margins that are within likely run-to-run variation, and by an internal inconsistency about which configuration produced the main results. The semantic-interpretation claim for the gates is plausible but not quantitatively validated. These issues are fixable, but they need to be addressed before the central claim can be accepted.

major comments (4)
  1. [§4.1 vs §4.3, Table 5] The manuscript is internally inconsistent about the architecture that produced the main results. Section 4.1 states that "4 DSA models and 4 DSS models are stacked to construct DSAN," and Fig. 2 depicts this stacked design. However, Section 4.3 and Table 5 compare one, two, and four DSA models and report that two DSA models give the best NMEip (6.24), four give 6.50, and one gives 6.27. The value 6.27 is exactly the NMEip reported for DSAT on the 300W Challenging subset in Table 1. This creates a three-way conflict: if the final model has four DSA modules, the main results are not for the described architecture; if it has two, the implementation details and figure are wrong; if it has one, Table 1 does not reflect the described DSAT. Since the paper's central claim is about the DSAT architecture as described, the authors must clarify the exact configuration used for each reported number and ensure the main tables, the ablation, and the architecture description agree.
  2. [Table 3 and §4.2 (COFW)] The COFW results mix two different normalization metrics in a single table and in the text. Table 3 reports an NMEip block in which DSAT's 4.74 is worse than LDDMM-Face's 4.54, and an NMEio block in which DSAT's 2.57 is the best. The sentence "The result of NMEip is 4.74, which is a huge improvement" is therefore false under the NMEip comparison; the improvement claim can only be made for NMEio, and even then the comparison must be against methods evaluated with the same metric. The authors should either report a single consistent metric for COFW or clearly separate the NMEio and NMEip comparisons and remove the misleading "huge improvement" statement.
  3. [§4.2, Tables 1, 2, and 4] Several claimed wins over state-of-the-art methods are extremely small and are reported without error bars, multiple trials, or significance tests. For example, on 300W Fullset DSAT reports 2.86 NMEio versus STAR's 2.87, on WFLW DSAT reports 4.12 versus SLPT's 4.14, and on AFLW DSAT reports 1.35 versus DTLD's 1.37. These differences are likely within run-to-run variation for heatmap regression models trained with different random seeds. The authors should provide variance estimates, standard deviations over repeated runs, or an accepted statistical comparison, and should temper the language "outperforms state-of-the-art" accordingly if the margins cannot be shown to be reliable.
  4. [§3.1, Eqs. (1)–(4)] The core semantic-gating assumption is not validated quantitatively. The DSA module derives binary gates from channel-wise adaptive-average-pooled values, additive Gaussian noise, a saturating sigmoid, and a threshold, but the paper does not demonstrate that the resulting gates track any learned notion of semantic relevance or that samples with similar gates are semantically similar beyond the qualitative examples in Figures 6 and 7. If the gating is effectively random with respect to sample content, the DSA mechanism reduces to stochastic channel dropout and the paper's central mechanism collapses. I recommend an additional analysis that measures, for example, gate agreement across augmented views of the same image, gate separation between annotated subsets (occlusion, pose, makeup), or the learned-correlation structure between the pooled vector and the gates, to support the semantic-interpretation claim.
minor comments (5)
  1. [§3.1, Eq. (4)] Equation (4) contains a typo: the text refers to "d??" instead of "dβ", and the sentence "dβ is a binary vector containing only 0 ones" should read "containing only 0s and 1s."
  2. [§4.1] The dataset name "ALFW" should be "AFLW" throughout the paper, including the first mention in Section 4.1.
  3. [§4.2 (WFLW)] The claim that DSAT "improves by 2% and 4% in the make-up Subset and Occlusion Subset" is ambiguous because the improvements are relative percentage changes, not absolute NME differences. Please state the absolute differences and clarify the comparison protocol.
  4. [§3.2] The statement that the CCA module "repeats the multi-head cross-channel attention for several times" is vague; Section 4.1 later specifies three CCA modules, but the architecture description should be consistent with the implementation details from the start.
  5. [§4.3] The ablation table reports only NMEip on 300W; reporting the same ablations for NMEio and for at least one other dataset would strengthen the conclusions about the DSA and DSS contributions.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the DSAT result is evaluated on external benchmarks and is not derived from its own assumptions.

full rationale

The paper's central claim is an empirical comparison: DSAT reports NME/FR values on the held-out test sets of 300W, WFLW, COFW and AFLW, using a standard L2 heatmap/boundary loss (Eq. 13) and the standard NME metric (Eq. 14). The DSA gate vector is learned end-to-end with a straight-through estimator rather than being fitted to the reported NME numbers; the DSS module is a learned attention module. The only self-citations are [17], [36], and [38] as related-work baselines or as descriptions of prior backbones, and none is invoked as evidence that DSAT's mechanism works. The observation that harder samples activate more channels is descriptive post-hoc analysis of the trained gates, not a prediction obtained from a fitted parameter. The manuscript's internal inconsistency about the number of DSA modules (Section 4.1 says four stacked modules, while Section 4.3/Table 5 reports two as best, and 6.27 for one matches Table 1) is a reproducibility/correctness concern, but it is not a circularity: the benchmark numbers are not defined in terms of the ablation configuration. Thus no load-bearing step reduces to its own input.

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

The central claim depends on several hyperparameters selected by ablation and on assumptions about the meaningfulness of channel gating and cross-scale attention fusion. Most are design choices rather than physical constants, and the most fragile premise is that raw pooled feature values encode semantic channel correlations.

free parameters (5)
  • Number of CCA modules per DSS = 3
    Ablation on 300W NMEip (Table 5): 3 modules gives 6.24 vs 6.41 for 4 modules; used in the final model.
  • Number of attention heads in CCA = 4
    Ablation on 300W NMEip: 4 heads gives 6.24 vs 6.41 for 2 heads and 6.38 for 8 heads.
  • Number and placement of DSA models = 2 (before first and second DSS)
    Table 5 says first-and-second is best (6.24) and 4 before each DSS is worse (6.50), contradicting the implementation text that says 4 DSA models are stacked.
  • Noise distribution for binary gates = Gaussian N(0,1)
    Section 3.1 adds unit Gaussian noise to pooled features for straight-through training; no ablation on noise scale is reported.
  • Learning rate schedule = 2.5e-4 initial, halved every 40k iterations, 100k total
    Implementation detail chosen without ablation; it directly affects the final NME values.
assumptions (5)
  • ad hoc to paper Channel-wise adaptive-average-pooled values (plus noise) reflect semantic relevance for sample partitioning.
    Section 3.1, Eqs. 1-4: no learned correlation model is described and no validation that gates correspond to semantic categories is provided.
  • domain assumption Improved SemHash straight-through estimator transfers from sequence models to face alignment without hurting gradient flow.
    Adopted from Ref. [27]; the paper provides no analysis of gate collapse or gradient behavior in this new setting.
  • ad hoc to paper Harder samples require and benefit from more activated channels.
    Figs. 1, 6, and 7 and Section 4.3; there is no formal definition of difficulty and no quantitative test, so the claim is descriptive.
  • domain assumption Cross-channel attention across scales reduces semantic gap and noise rather than adding noise.
    DSS design premise; evaluated only through end-to-end ablations, not measured directly.
  • domain assumption Published NME numbers from different papers are directly comparable when the normalization matches.
    SOTA comparisons in Tables 1-4 assume consistent evaluation protocols; the COFW table mixes NMEio and NMEip.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Precise Facial Landmark Detection by Dynamic Semantic Aggregation Transformer." pith.science (2026). https://pith.science/paper/ET3XT3QE

@misc{pith2026241200740,
  author       = {Pith},
  title        = {Pith review of: Precise Facial Landmark Detection by Dynamic Semantic Aggregation Transformer},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ET3XT3QE}},
  note         = {Machine review of arXiv:2412.00740}
}
read the original abstract

At present, deep neural network methods have played a dominant role in face alignment field. However, they generally use predefined network structures to predict landmarks, which tends to learn general features and leads to mediocre performance, e.g., they perform well on neutral samples but struggle with faces exhibiting large poses or occlusions. Moreover, they cannot effectively deal with semantic gaps and ambiguities among features at different scales, which may hinder them from learning efficient features. To address the above issues, in this paper, we propose a Dynamic Semantic-Aggregation Transformer (DSAT) for more discriminative and representative feature (i.e., specialized feature) learning. Specifically, a Dynamic Semantic-Aware (DSA) model is first proposed to partition samples into subsets and activate the specific pathways for them by estimating the semantic correlations of feature channels, making it possible to learn specialized features from each subset. Then, a novel Dynamic Semantic Specialization (DSS) model is designed to mine the homogeneous information from features at different scales for eliminating the semantic gap and ambiguities and enhancing the representation ability. Finally, by integrating the DSA model and DSS model into our proposed DSAT in both dynamic architecture and dynamic parameter manners, more specialized features can be learned for achieving more precise face alignment. It is interesting to show that harder samples can be handled by activating more feature channels. Extensive experiments on popular face alignment datasets demonstrate that our proposed DSAT outperforms state-of-the-art models in the literature.Our code is available at https://github.com/GERMINO-LiuHe/DSAT.

Figures

Figures reproduced from arXiv: 2412.00740 by the authors.

Figure 1
Figure 1. The ratios of activated channels for different samples. The harder the face image, the greater the ratio of activated channels. This indicates that our proposed DSAT could learn specialized features according to the difficulty estimation of the face images for achieving more precise face alignment. interest in the academic community, including face recognition [1, 2], face synthesis [3] and expression recognition [4… view at source ↗
Figure 2
Figure 2. The overall architecture of the proposed Dynamic Semantic Aggregation Transformer (DSAT). The proposed DSAT embeds [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. The architecture of the proposed Dynamic Semantic-Aware (DSA) model. The DSA model is able to partition all training [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: The network architecture of the DSS model. With a well-designed Cross-Channel Attention (CCA) module, the DSS model [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: The image shows the average number of DSA model channels activated on WLFW datasets. For di [PITH_FULL_IMAGE:figures/full_fig_p016_5.png]
Figure 6
Figure 6. Figure 6: Channel activation in different cases on 300W challenging subset. The DSA model activates different channels according to the estimated sample difficulty. The harder the sample, the more channels are activated [PITH_FULL_IMAGE:figures/full_fig_p017_6.png]
Figure 7
Figure 7. Figure 7: Channel activation of different samples in WFLW make-up subset. The heavy makeup face activates more channels than the light makeup one. will activate different channels according to the estimated sample difficulty. For example, in [PITH_FULL_IMAGE:figures/full_fig_p0…
Figure 8
Figure 8. Figure 8: Partial feature maps generated by DSAT with (first line) and without the DSS model (second line). We can find that by [PITH_FULL_IMAGE:figures/full_fig_p018_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

38 extracted references · 38 canonical work pages

  1. [1]

    Wright, A

    J. Wright, A. Y . Yang, A. Ganesh, S. Sastry, Y . Ma, Robust face recognition via sparse representation, IEEE Transactions on Pattern Analysis and Machine Intelligence 31 (2009) 210–227

    work page 2009

  2. [2]

    M. He, J. Zhang, S. Shan, X. Chen, Enhancing face recognition with detachable self-supervised bypass networks, IEEE Transactions on Image Processing 33 (2024) 1588–1599

    work page 2024

  3. [3]

    Z. Yang, J. Q. Liang, C. Fu, M. Luo, X. Zhang, Heterogeneous face recognition via face synthesis with identity-attribute disentanglement, IEEE Transactions on Information Forensics and Security 17 (2022) 1344–1358

    work page 2022

  4. [4]

    T. Liu, J. Li, J. Wu, L. Zhang, S. Zhao, J. Chang, J. Wan, Cross-domain facial expression recognition via disentangling identity representation., in: IJCAI, 2023, pp. 1213–1221

    work page 2023

  5. [5]

    Savchenko, L

    A. Savchenko, L. V . Savchenko, I. Makarov, Classifying emotions and engagement in online learn- ing based on a single facial expression recognition neural network, IEEE Transactions on A ffective Computing 13 (2022) 2132–2143

    work page 2022

  6. [6]

    Y . Zhao, Y . Liu, C. Shen, Y . Gao, S. Xiong, Mobilefan: Transferring deep hidden representation for face alignment, Pattern Recognit. 100 (2019) 107114

    work page 2019

  7. [7]

    T. Wu, K. Wang, C. Tang, J. Zhang, Di ffusion-based network for unsupervised landmark detection, Knowledge-Based Systems 292 (2024) 111627

    work page 2024

  8. [8]

    Y . Tian, D. Su, S. Li, Adaptive robust loss for landmark detection, Information Fusion 101 (2024) 102013

    work page 2024

Show all 38 references
  1. [9]

    J. M. Johnson, T. M. Khoshgoftaar, Survey on deep learning with class imbalance, Journal of Big Data 6 (2019) 1–54

    2019

  2. [10]

    Y . Han, G. Huang, S. Song, L. Yang, H. Wang, Y . Wang, Dynamic neural networks: A survey, IEEE Transactions on Pattern Analysis and Machine Intelligence 44 (2021) 7436–7456

    2021

  3. [11]

    Z. Xie, L. Jin, X. Du, X. Xiao, H. Li, S. Li, On generalized rmp scheme for redundant robot manip- ulators aided with dynamic neural networks and nonconvex bound constraints, IEEE Transactions on Industrial Informatics 15 (2019) 5172–5181

    2019

  4. [12]

    Cootes, C

    T. Cootes, C. J. Taylor, D. H. Cooper, J. Graham, Active shape models-their training and application, Comput. Vis. Image Underst. 61 (1995) 38–59. 20

    1995

  5. [13]

    Cootes, G

    T. Cootes, G. J. Edwards, C. J. Taylor, Active appearance models, in: European Conference on Com- puter Vision, 1998

    1998

  6. [14]

    C. Zhu, X. Li, J. Li, S. Dai, W. Tong, Reasoning structural relation for occlusion-robust facial landmark localization, Pattern Recognition 122 (2022) 108325

    2022

  7. [15]

    Liang, H

    J. Liang, H. Liu, H. Xu, D. Luo, Generalizable face landmarking guided by conditional face warping, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 2425–2435

    2024

  8. [16]

    Z. Zhou, H. Li, H. Liu, N. na Wang, G. Yu, R. Ji, Star loss: Reducing semantic ambiguity in facial landmark detection, 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2023) 15475–15484

    2023

  9. [17]

    J. Wan, J. Liu, J. Zhou, Z. Lai, L. Shen, H. Sun, P. Xiong, W. Min, Precise facial landmark detection by reference heatmap transformer, IEEE Transactions on Image Processing 32 (2023) 1966–1977

    2023

  10. [18]

    Cristinacce, T

    D. Cristinacce, T. Cootes, Feature detection and tracking with constrained local models, in: British Machine Vision Conference, 2006

    2006

  11. [19]

    Liu, Discriminative face alignment, IEEE Transactions on Pattern Analysis and Machine Intelli- gence 31 (2009) 1941–1954

    X. Liu, Discriminative face alignment, IEEE Transactions on Pattern Analysis and Machine Intelli- gence 31 (2009) 1941–1954

    2009

  12. [20]

    C. Lin, B. Zhu, Q. Wang, R. Liao, C. Qian, J. Lu, J. Zhou, Structure-coherent deep feature learning for robust face alignment, IEEE Transactions on Image Processing 30 (2021) 5313–5326

    2021

  13. [21]

    Z. Xu, B. Li, M. Geng, Y . Yuan, G. Yu, Anchorface: An anchor-based facial landmark detector across large poses, in: AAAI Conference on Artificial Intelligence, 2020

    2020

  14. [22]

    J. Xia, W. Qu, W.-F. Huang, J. Zhang, X. Wang, M. Xu, Sparse local patch transformer for robust face alignment and landmarks inherent relation learning, 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2022) 4042–4051

    2022

  15. [23]

    H. Yang, J. Lyu, P. Cheng, R. Tam, X. Tang, Lddmm-face: Large deformation di ffeomorphic metric learning for cross-annotation face alignment, Pattern Recognition (2024)

    2024

  16. [24]

    X. Wang, L. Bo, F. Li, Adaptive wing loss for robust face alignment via heatmap regression, 2019 IEEE/CVF International Conference on Computer Vision (ICCV) (2019) 6970–6980

    2019

  17. [25]

    J. Ma, J. Li, B. Du, J. Wu, J. Wan, Y . Xiao, Robust face alignment by dual-attentional spatial-aware capsule networks, Pattern Recognit. 122 (2022) 108297

    2022

  18. [26]

    Huang, H

    Y . Huang, H. Yang, C. Li, J. Kim, F. Wei, Adnet: Leveraging error-bias towards normal direction in face alignment, 2021 IEEE/CVF International Conference on Computer Vision (ICCV) (2021) 3060– 3070. 21

    2021

  19. [27]

    Kaiser, S

    L. Kaiser, S. Bengio, A. Roy, A. Vaswani, N. Parmar, J. Uszkoreit, N. Shazeer, Fast decoding in sequence models using discrete latent variables, in: International Conference on Machine Learning, ICML, 2018, pp. 2390–2399

    2018

  20. [28]

    H. Sun, Z. Luo, D. Ren, B. Du, L. Chang, J. Wan, Unsupervised multi-branch network with high- frequency enhancement for image dehazing, Pattern Recognition (2024)

    2024

  21. [29]

    W. Wu, C. Qian, S. Yang, Q. Wang, Y . Cai, Q. Zhou, Look at boundary: A boundary-aware face alignment algorithm, 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition (2018) 2129–2138

    2018

  22. [30]

    K ¨ostinger, P

    M. K ¨ostinger, P. Wohlhart, P. M. Roth, H. Bischof, Annotated facial landmarks in the wild: A large- scale, real-world database for facial landmark localization, 2011 IEEE International Conference on Computer Vision Workshops (ICCV Workshops) (2011) 2144–2151

    2011

  23. [31]

    Sagonas, G

    C. Sagonas, G. Tzimiropoulos, S. Zafeiriou, M. Pantic, 300 faces in-the-wild challenge: The first facial landmark localization challenge, 2013 IEEE International Conference on Computer Vision Workshops (2013) 397–403

    2013

  24. [32]

    X. P. Burgos-Artizzu, P. Perona, P. Doll ´ar, Robust face landmark estimation under occlusion, 2013 IEEE International Conference on Computer Vision (2013) 1513–1520

    2013

  25. [33]

    H. Li, Z. Guo, S.-M. Rhee, S. J. Han, J.-J. Han, Towards accurate facial landmark detection via cas- caded transformers, 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2022) 4166–4175

    2022

  26. [34]

    C. Zhu, X. Li, J. Li, G. Ding, W. Tong, Spatial-temporal knowledge integration: Robust self-supervised facial landmark tracking, Proceedings of the 28th ACM International Conference on Multimedia (2020)

    2020

  27. [35]

    K. Yin, V . Rao, R. Jiang, X. Liu, P. Aarabi, D. B. Lindell, Sce-mae: Selective correspondence en- hancement with masked autoencoder for self-supervised landmark estimation, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 1313–1322

    2024

  28. [36]

    J. Wan, Z. Lai, J. Liu, J. Zhou, C. Gao, Robust face alignment by multi-order high-precision hourglass network, IEEE Transactions on Image Processing 30 (2020) 121–133

    2020

  29. [37]

    J. P. Robinson, Y . Li, N. Zhang, Y . R. Fu, S. Tulyakov, Laplace landmark localization, 2019 IEEE/CVF International Conference on Computer Vision (ICCV) (2019) 10102–10111

    2019

  30. [38]

    P. Gao, K. Lu, J. Xue, J. Lyu, L. Shao, A facial landmark detection method based on deep knowledge transfer, IEEE Transactions on Neural Networks and Learning Systems 34 (2021) 1342–1353. 22

    2021

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.