Pith. sign in

REVIEW 4 major objections 5 minor 27 references

Boosting Sclera Segmentation through Semi-supervised Learning with Fewer Labels

T0 review · 4 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read A two-stage semi-supervised framework with domain-specific and spatial-transform augmentations attains accurate sclera segmentation from as few as four labeled images, beating the original SSL method and most prior models.

desk verdict The SSL advantage is never isolated from the backbone change, and without a supervised-only control the paper's central claim is not supported; still, the new dataset and sensible SSL extension make it worth reviewing. read the letter →

arxiv 2501.07750 v1 pith:JBI243VH submitted 2025-01-13 cs.CV

Guanjun Wang , Lu Wang , Ning Niu , Qiaoyi Yao , Yixuan Wang , Sufen Ren , Shengchao Chen This is my paper · ORCID
classification cs.CV
keywords sclerasegmentationsemi-supervisedlearningconsistencyregularizationdataaugmentationCLAHEU2Netmedicalimagelabel-efficient
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 claims that sclera segmentation, a step needed for eye-disease diagnosis and biometric identification, can be trained from a handful of labeled images instead of hundreds. Its framework pairs semi-supervised learning with two kinds of augmentation: domain-specific adjustments (CLAHE and gamma correction) that stabilize predictions under lighting variation, and spatial transforms (rotation and translation) whose inverse-averaged predictions serve as self-supervised targets for unlabeled data. The segmentation backbone is an enlarged U2Net with seven encoder and six decoder stages. On the authors' new eye-diagnosis dataset, the method reports 87.94% mIoU with only four labeled examples; on UBIRIS.v2 and SBVPI it matches or beats prior models that were trained on many more labels. If correct, this makes accurate eye segmentation accessible where expert annotations are scarce.

What carries the argument

The load-bearing machinery is the two-stage consistency framework together with the enlarged U2Net. In the first stage (SSLD), each image is augmented $k$ times with CLAHE and gamma correction; the model's softmax predictions on the augmented copies are averaged to form guessed labels for unlabeled data, and an L2 loss ($L_u$) enforces agreement with those guesses. In the second stage (SSL-SS), a spatial transform $T$ (small rotations and translations) is applied before prediction, and the inverse transforms $T^{-1}$ are applied to the predictions before averaging per Eq. (6), so the guessed labels are consistent under geometric perturbation. The supervised loss $L_s$ combines cross-entropy, boundary-aware, dice, and surface losses with a schedule that weights dice first and surface loss later. The segmentation backbone is U2Net expanded from six to seven encoder levels and five to six decoder levels, with RSU blocks and dilated RSU-4F in the deepest stages and a Saliency Graph Fusion Module that fuses seven side maps into the final saliency map. This combination is what the paper credits for the rapid convergence and label efficiency.

What would settle it

During training, compute the agreement between the guessed labels and the held-out ground truth on a small labeled subset the model never trains on. If that agreement falls below the accuracy of a model trained on the supervised loss alone while the unsupervised loss keeps decreasing, the pseudo-label loop is teaching systematic errors; likewise, corrupting the ground-truth masks with a known pixel shift and observing whether the unsupervised losses amplify the shift faster than the supervised loss corrects it would isolate the same failure.

Watch

Extended reading notes

Core claim

The paper's central claim is that its SSL framework, combining domain-specific augmentations in a first stage (SSLD), a second stage (SSL-SS) that enforces consistency across spatial transformations by averaging inverse-transformed predictions, and an improved U2Net backbone, attains high segmentation accuracy with as few as four labeled images, where the original SSL method stays underfitted within the same 100 training epochs. The authors report that with the same RITnet segmentation network, their method's test mIoU at four labels already exceeded the original method's mIoU at 96 labels. They also state that neither SSL variant with RITnet segmented the sclera effectively, which is why they moved to an enlarged U2Net as the backbone. On their dataset, the proposed method reaches 87.94% mIoU at $X_l = 4$ and 89.90% at $X_l = 96$; on UBIRIS.v2, 72 labeled examples yield 84.60% mIoU, surpassing most earlier models trained with 120 labeled examples; on SBVPI, 500 labeled examples yield 91.77% mIoU, above ScleraSegNet's result with 734 labels. The paper additionally contributes a manually annotated eye-diagnosis dataset of about 800 images from over 100 patients.

Load-bearing premise

The method assumes that the pseudo-labels formed by averaging the model's own predictions across augmentations and inverse-transformed views are reliable enough to train on; if those guesses carry systematic errors, the unsupervised losses will reinforce them instead of correcting them.

Editorial extensions

If this is right

  • Sclera segmentation becomes feasible where expert annotations are scarce: four labeled images plus unlabeled data yield 87.94% mIoU on the authors' dataset.
  • Label efficiency transfers across image domains: at 72 labeled examples the method reaches 84.60% mIoU on UBIRIS.v2, exceeding most prior methods trained on 120 labeled examples.
  • On the high-resolution SBVPI dataset, 500 labeled examples give 91.77% mIoU, surpassing ScleraSegNet, which used 734 labeled examples.
  • The spatial-consistency stage (SSL-SS) appears to be what lifts training out of the underfitting that the original SSL method showed within 100 epochs.
  • The new eye-diagnosis dataset of roughly 800 manually annotated images provides a benchmark for label-scarce sclera segmentation, with gaze directions that stress occlusion and illumination.

Reading between the lines

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

  • The consistency mechanism in Eq. (6) is a generic spatial regularizer with no sclera-specific parts, so the same two-stage recipe could be lifted to other anatomical segmentation tasks such as iris, pupil, or conjunctival vessels.
  • Because the guessed labels are unthresholded softmax averages, a natural extension the paper does not explore is confidence weighting or thresholding on pseudo-labels to suppress low-quality guesses.
  • The dataset's multiple gaze directions suggest evaluating segmentation under extreme gaze and eyelid occlusion separately; the paper reports only aggregate metrics, leaving per-condition performance open.
  • Ablating the SSLD and SSL-SS stages separately would reveal which stage carries the label efficiency, since the paper reports only the combined method.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper proposes a semi-supervised learning framework for sclera segmentation that combines domain-specific augmentations (CLAHE, gamma correction), a two-stage SSL procedure (SSLD and SSL-SS) based on consistency across augmentations and inverse spatial transformations, and an enhanced seven-level U2Net backbone. The authors introduce a new eye-diagnosis dataset of roughly 800-1000 manually annotated images and evaluate the method on this dataset, UBIRIS.v2, and SBVPI using only 4 to 96 labeled images. The central claim is that the proposed SSL framework achieves accurate sclera segmentation with very few labels, outperforming the original SSL method and most prior fully supervised sclera segmentation models.

Significance. If the claims are supported, the work would be a useful contribution to low-label medical image segmentation and sclera biometrics: the new dataset is a potential community resource, the domain-specific augmentations are sensible, and the experiments span three datasets with systematically varied label counts. The paper also gives credit for attempting a real-world diagnostic dataset rather than relying only on public benchmarks. However, the experimental design currently conflates the SSL machinery with the choice of segmentation backbone, and it lacks the supervised-only and same-backbone controls needed to attribute the reported gains to semi-supervised learning. The significance of the method is therefore not yet established by the evidence presented.

major comments (4)
  1. [§4.3, Table 2 and Fig. 6-8] The claimed SSL advantage is never isolated. The comparison to the original SSL method (Chaudhary et al., 2021) uses RITnet as the backbone for both methods, and the authors state that 'neither method effectively segmented the sclera' with RITnet. The proposed method is then evaluated with the enhanced U2Net, but the original SSL framework is never run with the enhanced U2Net. Consequently, the large gap between 'U2Net' and 'Proposed Method' in Table 2 could be due to the deeper seven-level architecture rather than to the SSLD/SSL-SS pseudo-labeling or the domain-specific augmentations. A supervised-only baseline is also missing: there is no experiment training the enhanced U2Net on the same labeled images while omitting the unsupervised losses Lu and Lss. Without these two controls, the contribution of the unlabeled data, which is the defining element of the claimed SSL boost, cannot be assessed.
  2. [§4.3, Table 4 and Table 3] The comparisons against prior work are not conducted under identical training conditions. Table 3 reports results of fully supervised methods trained with larger label counts (e.g., 120 labeled UBIRIS.v2 images for ScleraSegNet, 734 for SBVPI), while Table 4 reports the proposed method at various Xl values. The claim that Xl=72 on UBIRIS.v2 'surpasses the performance of most models listed in Table 3' is therefore not a head-to-head comparison, since the backbone, training schedule, loss terms, and number of labels all differ. The paper should provide same-protocol comparisons, at least for the main baselines, and should also report results over multiple random seeds with standard deviations or error bars, since the reported improvements are often small (e.g., 87.94 vs 86.45 mIoU for Xl=4 in Table 2).
  3. [§4.1, Table 1 and Abstract/Contributions] The dataset description is internally inconsistent and this is load-bearing for the evaluation. The abstract and contributions state 'approximately 800 images from over 100 patients,' but §4.1 says 'about 1000 images were used,' and Table 1 reports 700 training, 200 validation, and 100 test images, which sums to exactly 1000. The authors should clarify the true dataset size, report the number of subjects and images precisely, and state whether all experiments use the same train/validation/test split. In addition, since all evaluation depends on manual annotations, details on annotator expertise and inter-annotator agreement should be provided.
  4. [§4.2 and §3.3.1] Several hyperparameters required to reproduce Equations (5), (6), and (8) are missing or incompletely specified. The number of augmented copies k in Eqs. (5)-(6) is not defined; the unsupervised loss schedules are described only by slopes (0.02 and 0.002 per epoch) without their initial values and saturation behavior; and the sentence describing the augmentation probabilities ends with 'with probabilities of 50' in §4.2, leaving p1 and p2 unspecified. These values are needed for any independent verification of the reported results, and they are among the free parameters listed in the method.
minor comments (5)
  1. [Throughout] There are numerous typos and citation errors, including 'paer' in §4.1, 'n the initial phase' in §3.2, and 'Rot et al.(Ronneberger et al., 2015)' and 'Lucio et al.(Ronneberger et al., 2015)' in Section 1.2, which cite the wrong reference.
  2. [§4.3, Figures 6-8] The subcaptions in Figures 6-8 all use the same label '(a)' for different numbers of labeled images; each panel should be uniquely identified.
  3. [§3.3.1, Eqs. (5)-(6)] The notation k is introduced as the number of augmented copies but is never explicitly defined in the text; please define it before Eq. (5) and state its value in the experiments.
  4. [§4.1, Eq. (10)] The text 'is the result of the segmentation of ground true' should read 'ground truth', and the recall formula should be written with explicit set intersection/union or TP/FP/FN notation consistently.
  5. [§4.2] The sentence about the T transform ends with 'with probabilities of 50' and is incomplete; the values of p1 and p2 and the exact rotation/translation ranges should be given.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the reported segmentation metrics are evaluated on held-out ground truth and the method is benchmarked against external datasets; the SSL pseudo-labeling is a training objective, not a derivation of the reported results.

full rationale

The paper's central claim is empirical: the proposed SSL framework with domain-specific augmentations and an enhanced U2Net achieves high mIoU on held-out test sets (Tables 2 and 4). The unlabeled-data losses Lu and Lss use the model's own averaged or inverse-transformed predictions as guessed labels (Eqs. 5-6), but this is a standard self-training/consistency objective, not a circular derivation: the evaluation metrics in Eqs. 9-12 are computed against manually annotated ground-truth masks that are independent of the training-time pseudo-labels. The method is also tested on two public datasets, UBIRIS.v2 and SBVPI, and compared with prior external methods (Tables 3-4), making the performance claims externally falsifiable. The self-citations in the introduction and related work are contextual and not load-bearing; the SSL framework itself is attributed to the external work of Chaudhary et al. (2021). The lack of a supervised-only control and the backbone change between the RITnet comparison and the final U2Net comparison are validity concerns, but they are not circularity: no equation or fitted parameter is renamed as a prediction, and the reported gains are not forced by construction. Therefore, no significant circularity is present.

Assumptions & free parameters 9 free parameters · 4 assumptions · 0 invented entities

The paper introduces no new physical or conceptual entities. Its central claim depends on several hand-chosen hyperparameters and domain assumptions about the SSL consistency premise, pseudo-label quality, annotation correctness, and augmentation safety.

free parameters (9)
  • Unsupervised loss ramp slope (lambda_u) = 0.02 per epoch
    Linear weight increase for the unlabeled loss term in Eq. (8), chosen by the authors.
  • Self-supervised loss ramp slope (lambda_ss) = 0.002 per epoch
    Linear weight increase for the self-supervised loss term in Eq. (8), chosen by the authors.
  • Alpha schedule for supervised loss = epoch/100
    Weight balancing dice loss and surface loss in Eq. (7).
  • CLAHE clipping limits = (1.0, 1.2, 1.5, 1.5, 1.5, 2.0)
    Clipping thresholds for contrast-limited adaptive histogram equalization at the six encoder levels.
  • CLAHE grid sizes = (2, 4, 8, 8, 8, 16)
    Tile grid sizes for CLAHE at the six encoder levels.
  • Gamma correction range = 0.8 to 1.2 in steps of 0.05
    Random gamma values for domain-specific augmentation.
  • Rotation range for transform T = -5 to 5 degrees
    Range of random rotations applied to unlabeled images in the self-supervised stage.
  • Translation range for transform T = -20 to 20 pixels
    Range of random translations applied to unlabeled images.
  • Spatial transform probabilities p1, p2
    Probabilities for applying rotation and translation; the text cuts off at '50' and does not specify the values.
assumptions (4)
  • domain assumption Predictions of a model should be stable under transformations of the input (consistency assumption).
    This is the inductive bias underlying the unsupervised losses in Section 3.3.1; if it does not hold for sclera images, the pseudo-labels are unreliable.
  • domain assumption Guessed labels generated by the model on unlabeled data are informative for training.
    The method relies on self-training with pseudo-labels; incorrect pseudo-labels could reinforce errors.
  • domain assumption Manual annotations of the new dataset are accurate.
    All quantitative results depend on ground-truth masks annotated under physician guidance.
  • domain assumption CLAHE and gamma correction preserve sclera structure and do not distort labels.
    Domain-specific augmentations are applied to both labeled and unlabeled data; if they alter anatomical consistency, the learned mapping degrades.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Boosting Sclera Segmentation through Semi-supervised Learning with Fewer Labels." pith.science (2026). https://pith.science/paper/JBI243VH

@misc{pith2026250107750,
  author       = {Pith},
  title        = {Pith review of: Boosting Sclera Segmentation through Semi-supervised Learning with Fewer Labels},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JBI243VH}},
  note         = {Machine review of arXiv:2501.07750}
}
read the original abstract

Sclera segmentation is crucial for developing automatic eye-related medical computer-aided diagnostic systems, as well as for personal identification and verification, because the sclera contains distinct personal features. Deep learning-based sclera segmentation has achieved significant success compared to traditional methods that rely on hand-crafted features, primarily because it can autonomously extract critical output-related features without the need to consider potential physical constraints. However, achieving accurate sclera segmentation using these methods is challenging due to the scarcity of high-quality, fully labeled datasets, which depend on costly, labor-intensive medical acquisition and expertise. To address this challenge, this paper introduces a novel sclera segmentation framework that excels with limited labeled samples. Specifically, we employ a semi-supervised learning method that integrates domain-specific improvements and image-based spatial transformations to enhance segmentation performance. Additionally, we have developed a real-world eye diagnosis dataset to enrich the evaluation process. Extensive experiments on our dataset and two additional public datasets demonstrate the effectiveness and superiority of our proposed method, especially with significantly fewer labeled samples.

Figures

Figures reproduced from arXiv: 2501.07750 by the authors.

Figure 1
Figure 1. Overall structure of our method 3.2 DATA AUGMENTATION To overcome the challenge of insufficient data, data augmentation is the most common solution (Xu et al., 2023; Chen et al., 2022b;c;d). We incorporated domain-specific and transformation-based augmentation techniques into our semi-supervised learning (SSL) framework. Specifically, domain￾specific augmentation employs methods such as Contrast Limited Adaptive His… view at source ↗
Figure 2
Figure 2. Structure of the SSL framework 3.3.1 SSL FRAMEWORK The SSL approach used, as shown in Fig.2, is divided into two stages. The first stage involves SSL with domain-specific enhancements (SSLD), while the second stage utilizes SSL with self￾supervised learning (SSL-SS). Compared to the first stage, the SSL-SS stage introduces a more effective regularization strategy by generating labels through a self-supervised mechan… view at source ↗
Figure 3
Figure 3. Improved U2Net architecture outperforming 20 other leading methods in both qualitative and quantitative assessments(Qin et al., 2020). U2Net comprises three primary components: a six-level encoder, a five-level decoder, and a saliency graph fusion model integrated with both the decoder and the final level of the encoder. To enhance image feature extraction, we expanded the encoder to seven levels and the decoder to … view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Eye diagnostic instrument We conducted comparative experiments with other networks using two public datasets, UBIRIS.v2 (Proenc¸a et al., 2009) and SBVPI (Rot et al., 2018). UBIRIS.v2, initially developed for iris recog￾nition under less constrained conditions, compris…
Figure 5
Figure 5. Figure 5: Example images and corresponding sclera segmentation ground truths [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]
Figure 6
Figure 6. Figure 6: The results of the evaluation metrics of the two methods [PITH_FULL_IMAGE:figures/full_fig_p010_6.png]
Figure 7
Figure 7. Figure 7: Some segmentation results of different SSL method using RITnet as segmentation net [PITH_FULL_IMAGE:figures/full_fig_p011_7.png]
Figure 8
Figure 8. Figure 8: Some segmentation results of our SSL method using different segmentation networks [PITH_FULL_IMAGE:figures/full_fig_p013_8.png]
Figure 9
Figure 9. Figure 9: Sample images of the segmentation results of using proposed SSL method on UBIRIS.v2 [PITH_FULL_IMAGE:figures/full_fig_p014_9.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

27 extracted references · 22 canonical work pages

  1. [1]

    Federated Distillation for Medical Image Classification: Towards Trustworthy Computer-Aided Diagnosis

    Shengchao Chen, Sufen Ren, Guanjun Wang, Mengxing Huang, and Chenyang Xue. Interpretable cnn-multilevel attention transformer for rapid recognition of pneumonia from chest x-ray images. IEEE Journal of Biomedical and Health Informatics, 28(2):753–764, 2023a. Sufen Ren, Yule Hu, Shengchao Chen, and Guanjun Wang. Federated distillation for medical image cla...

    work page Pith review arXiv 2023

  2. [5]

    Deep multi-class eye segmentation for ocular biometrics

    Peter Rot, ˇZiga Emerˇsiˇc, Vitomir Struc, and Peter Peer. Deep multi-class eye segmentation for ocular biometrics. In 2018 IEEE international work conference on bioinspired intelligence (IWOBI) , pages 1–8. IEEE,

    work page 2018

  3. [7]

    Tempee: Temporal-spatial parallel transformer for radar echo extrapolation beyond auto-regression

    Shengchao Chen, Ting Shu, Huan Zhao, Guo Zhong, and Xunlai Chen. Tempee: Temporal-spatial parallel transformer for radar echo extrapolation beyond auto-regression. IEEE Transactions on Geoscience and Remote Sensing, 2023c. Wolfgang Fuhl, Thiago Santini, Gjergji Kasneci, and Enkelejda Kasneci. Pupilnet: Convolutional neural networks for robust pupil detect...

    arXiv

  4. [11]

    Less is more: Ensemble Learning for Retinal Disease Recognition Under Limited Resources

    Jiahao Wang, Hong Peng, Shengchao Chen, and Sufen Ren. Less is more: Ensemble learning for retinal disease recognition under limited resources. arXiv preprint arXiv:2402.09747, 2024b. Shengchao Chen, Ting Shu, Huan Zhao, and Yuan Yan Tang. Mask-cnn-transformer for real-time multi-label weather recognition. Knowledge-Based Systems, 278:110881, 2023e. Reza ...

    work page Pith review arXiv

  5. [17]

    Quality fusion based multimodal eye recognition

    Zhi Zhou, Eliza Yingzi Du, Craig Belcher, N Luke Thomas, and Edward J Delp. Quality fusion based multimodal eye recognition. In 2012 IEEE International Conference on Systems, Man, and Cybernetics (SMC), pages 1297–1302. IEEE,

    work page 2012

  6. [18]

    Sclera recog- nition using dense-sift

    Abhijit Das, Umapada Pal, Miguel Angel Ferrer Ballester, and Michael Blumenstein. Sclera recog- nition using dense-sift. In 2013 13th International Conference on Intellient Systems Design and Applications, pages 74–79. IEEE,

    work page 2013

  7. [19]

    A new efficient and adaptive sclera recognition system

    Abhijit Das, Umapada Pal, Miguel Angel Ferrer Ballester, and Michael Blumenstein. A new efficient and adaptive sclera recognition system. In 2014 IEEE Symposium on Computational Intelligence in Biometrics and Identity Management (CIBIM), pages 1–8. IEEE,

    work page 2014

  8. [21]

    Enhanced segmen- tation and complex-sclera features for human recognition with unconstrained visible-wavelength imaging

    Sinan Alkassar, Wai Lok Woo, Satnam Singh Dlay, and Jonathon A Chambers. Enhanced segmen- tation and complex-sclera features for human recognition with unconstrained visible-wavelength imaging. In 2016 International Conference on Biometrics (ICB), pages 1–8. IEEE,

    work page 2016

Show all 27 references
  1. [22]

    Fabry-perot interferometric sensor demodulation system utilizing multi-peak wavelength tracking and neural network algorithm

    Shengchao Chen, Feifan Yao, Sufen Ren, Jianli Yang, Qian Yang, Shuyu Yuan, Guanjun Wang, and Mengxing Huang. Fabry-perot interferometric sensor demodulation system utilizing multi-peak wavelength tracking and neural network algorithm. Optics express, 30(14):24461–24480, 2022b....

    arXiv

  2. [24]

    Semi-supervised medical image classification with global latent mixing

    Prashnna Kumar Gyawali, Sandesh Ghimire, Pradeep Bajracharya, Zhiyuan Li, and Linwei Wang. Semi-supervised medical image classification with global latent mixing. In Medical Image Com- puting and Computer Assisted Intervention–MICCAI 2020: 23rd International Conference, Lima, ...

    2020

  3. [25]

    Revisiting self-supervised visual repre- sentation learning

    Alexander Kolesnikov, Xiaohua Zhai, and Lucas Beyer. Revisiting self-supervised visual repre- sentation learning. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 1920–1929,

    1920

  4. [26]

    The 2005 pascal visual object classes challenge

    Mark Everingham, Andrew Zisserman, Christopher KI Williams, Luc Van Gool, Moray Allan, Christopher M Bishop, Olivier Chapelle, Navneet Dalal, Thomas Deselaers, Gyuri Dork ´o, et al. The 2005 pascal visual object classes challenge. In Machine Learning Challenges. Evaluating Pre...

    2005

  5. [27]

    Adam: A method for stochastic optimization

    Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980,

    arXiv

  6. [2006]

    A texture-based neural network classifier for biometric iden- tification using ocular surface vasculature

    Reza Derakhshani and Arun Ross. A texture-based neural network classifier for biometric iden- tification using ocular surface vasculature. In 2007 International Joint Conference on Neural Networks, pages 2982–2987. IEEE,

    2007

  7. [2007]

    Enhancement and registration schemes for matching conjunctival vasculature

    Simona Crihalmeanu, Arun Ross, and Reza Derakhshani. Enhancement and registration schemes for matching conjunctival vasculature. In Advances in Biometrics: Third International Conference, ICB 2009, Alghero, Italy, June 2-5,

    2009

  8. [2009]

    A new approach for sclera vein recognition

    N Luke Thomas, Yingzi Du, and Zhi Zhou. A new approach for sclera vein recognition. In Mo- bile Multimedia/Image Processing, Security, and Applications 2010, volume 7708, pages 38–47. SPIE,

    2010

  9. [2010]

    Extracting sclera features for cancelable identity verification

    Kangrok Oh and Kar-Ann Toh. Extracting sclera features for cancelable identity verification. In 2012 5th IAPR International Conference on Biometrics (ICB), pages 245–250. IEEE,

    2012

  10. [2012]

    Dynamic multi- scale fusion generative adversarial network for radar image extrapolation

    Shengchao Chen, Ting Shu, Huan Zhao, Qilin Wan, Jincan Huang, and Cailing Li. Dynamic multi- scale fusion generative adversarial network for radar image extrapolation. IEEE Transactions on Geoscience and Remote Sensing, 60:1–11, 2022a. 15 Preprint Version for Submission Shengc...

    arXiv

  11. [2013]

    A comprehensive sciera image quality measure

    16 Preprint Version for Submission Zhi Zhou, Eliza Y Du, and N Luke Thomas. A comprehensive sciera image quality measure. In 2010 11th International Conference on Control Automation Robotics & Vision , pages 638–643. IEEE,

    2010

  12. [2014]

    Sclera vein identification in real time using single board computer

    KV Delna, KA Sneha, and RP Aneesh. Sclera vein identification in real time using single board computer. In 2016 International Conference on Next Generation Intelligent Systems (ICNGIS) , pages 1–5. IEEE,

    2016

  13. [2015]

    Scleral buckle surgery for primary retinal detachment without posterior vitreous detachment

    Jila Noori, Richard A Bilonick, and Andrew W Eller. Scleral buckle surgery for primary retinal detachment without posterior vitreous detachment. Retina, 36(11):2066–2071,

  14. [2016]

    Semi-supervised learn- ing by disentangling and self-ensembling over stochastic latent space

    Prashnna Kumar Gyawali, Zhiyuan Li, Sandesh Ghimire, and Linwei Wang. Semi-supervised learn- ing by disentangling and self-ensembling over stochastic latent space. InMedical Image Comput- ing and Computer Assisted Intervention–MICCAI 2019: 22nd International Conference, Shen- ...

    2019

  15. [2017]

    Temporal ensembling for semi-supervised learning

    Samuli Laine and Timo Aila. Temporal ensembling for semi-supervised learning. arXiv preprint arXiv:1610.02242,

    arXiv

  16. [2018]

    U-net: Convolutional networks for biomed- ical image segmentation

    Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-net: Convolutional networks for biomed- ical image segmentation. In Medical Image Computing and Computer-Assisted Intervention– MICCAI 2015: 18th International Conference, Munich, Germany, October 5-9, 2015, Proceed- ings, ...

    2015

  17. [2019]

    An image enhancement method based on gamma correction

    Xu Guan, Su Jian, Pan Hongda, Zhang Zhiguo, and Gong Haibin. An image enhancement method based on gamma correction. In 2009 Second international symposium on computational intelli- gence and design, volume 1, pages 60–63. IEEE,

    2009

  18. [2021]

    Sclera- transfuse: Fusing swin transformer and cnn for accurate sclera segmentation

    Haiqing Li, Caiyong Wang, Guangzhe Zhao, Zhaofeng He, Yunlong Wang, and Zhenan Sun. Sclera- transfuse: Fusing swin transformer and cnn for accurate sclera segmentation. In 2023 IEEE International Joint Conference on Biometrics (IJCB), pages 1–8. IEEE,

    2023

  19. [2023]

    Foundation mod- els for weather and climate data understanding: A comprehensive survey

    Shengchao Chen, Guodong Long, Jing Jiang, Dikai Liu, and Chengqi Zhang. Foundation mod- els for weather and climate data understanding: A comprehensive survey. arXiv preprint arXiv:2312.03014, 2023b. Shengchao Chen, Ting Shu, Huan Zhao, Jiahao Wang, Sufen Ren, and Lina Yang. F...

    arXiv

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.