pith. sign in

arxiv: 2604.22824 · v2 · submitted 2026-04-19 · 💻 cs.CV · cs.AI

WeatherSeg: Weather-Robust Image Segmentation using Teacher-Student Dual Learning and Classifier-Updating Attention

Zhang Zhang , Yifeng Zeng , Houshi Jiang , Yinghui Pan This is my paper

Pith reviewed 2026-05-10 06:38 UTC · model grok-4.3

classification 💻 cs.CV cs.AI
keywords semantic segmentationadverse weatherteacher-student learningsemi-supervised learningautonomous drivingattention mechanismweather robustnessimage segmentation
0
0 comments X

The pith

WeatherSeg delivers accurate semantic segmentation in rain, fog, clouds and clear conditions using a dual teacher-student model with dynamic classifier attention.

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

The paper introduces WeatherSeg as a semi-supervised segmentation framework that tackles the drop in performance autonomous driving systems experience when weather reduces visibility. It pairs a Dual Teacher-Student Weight-Sharing Model that distills knowledge from degraded images with a Classifier Weight Updating Attention Mechanism that tunes classifier weights according to detected environmental conditions. The approach is meant to improve both accuracy and robustness while lowering the need for expensive manual labels. A reader would care because current vision systems lose reliability precisely when safe driving decisions matter most.

Core claim

WeatherSeg integrates a Dual Teacher-Student Weight-Sharing Model that enables knowledge distillation from weather-affected images and a Classifier Weight Updating Attention Mechanism that dynamically adjusts classifier weights based on environmental attributes, allowing it to significantly outperform baseline models in both accuracy and robustness across clear, rainy, cloudy, and foggy scenarios.

What carries the argument

The Dual Teacher-Student Weight-Sharing Model combined with the Classifier Weight Updating Attention Mechanism, which transfers knowledge across weather-degraded images while adapting classifier weights to current conditions.

If this is right

  • The framework achieves higher segmentation accuracy than baselines under every tested weather condition.
  • Performance remains strong in clear weather while rising in rain, clouds, and fog.
  • Annotation effort drops because the method works effectively with limited labeled examples.
  • The same architecture supplies a practical route to all-weather perception for autonomous driving.

Where Pith is reading between the lines

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

  • The same weight-sharing and attention pattern could be applied to video streams to keep segmentation stable across changing weather during a drive.
  • Analogous dual-learning setups might improve robustness in other variable-quality imaging domains such as underwater or satellite scenes.
  • Real-time deployment tests on vehicles would show whether the attention updates keep pace with sudden weather shifts.

Load-bearing premise

The premise that weight sharing between teacher and student plus attention-driven classifier updates can reliably extract useful knowledge from degraded images without introducing new biases or needing undisclosed weather-specific adjustments.

What would settle it

Evaluation on a fresh mixed-weather autonomous-driving test set showing no statistically significant accuracy or robustness gain over a standard semi-supervised baseline such as Mean Teacher would falsify the central performance claim.

Figures

Figures reproduced from arXiv: 2604.22824 by Houshi Jiang, Yifeng Zeng, Yinghui Pan, Zhang Zhang.

Figure 1
Figure 1. Figure 1: WeatherSeg: The whole network architecture of the dual teacher-student with classifier weight [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: WeatherSeg: The first part is DTSWSM, whose primary role is to provide a stable foundational feature representation for the subsequent Classifier Weight Updating Attention Mechanism (CWUAM) dynamic weighting through parameter sharing and feature unification. CWUAM sharpens the basic feature output of the first part, accurately classifies the category importance, and represents the stable features as task-a… view at source ↗
Figure 3
Figure 3. Figure 3: A. Dual Teacher-Student Weight-Sharing Model(DTSWSM): It uses a shared encoder to extract [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: B. Classifier Weight Updating Attention Mechanism: By integrating the features of dual teachers and [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: A vehicle on the road, shown from a third-person view in 6 weather conditions in 3 time zones. For [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Wet Cloudy Night - DTSWSM Segmentation & Introduced CWUAM Predicted Results - SegNet,FCN,Unet and CWUAM with SegNet, U-Net, and FCN across various adverse weather conditions (Section A, Ta￾ble IV), we draw the following conclusions: Our analysis demonstrates the consistent superiority of the CWUAM attention training mechanism, which surpasses DTSWSM by an average mIoU of 0.83% across all base networks. We … view at source ↗
Figure 7
Figure 7. Figure 7: Loss Comparsion: Compare the Pseudo-label loss, Consistency loss and total values of DTSWSM and [PITH_FULL_IMAGE:figures/full_fig_p008_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Loss Comparsion - Ours & Current Methods: (a) [PITH_FULL_IMAGE:figures/full_fig_p013_8.png] view at source ↗
read the original abstract

WeatherSeg, an advanced semi-supervised segmentation framework, addresses autonomous driving's environmental perception challenges in adverse weather while reducing annotation costs. This framework integrates a Dual Teacher-Student Weight-Sharing Model (DTSWSM) that enables knowledge distillation from weather-affected images, and a Classifier Weight Updating Attention Mechanism (CWUAM) that dynamically adjusts classifier weights based on environmental attributes. Comprehensive evaluations demonstrate that WeatherSeg significantly outperforms baseline models in both accuracy and robustness across various weather conditions, including clear, rainy, cloudy, and foggy scenarios, establishing it as an effective solution for all-weather semantic segmentation in autonomous driving and related applications.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper presents WeatherSeg, a semi-supervised segmentation framework for weather-robust image segmentation in autonomous driving. It introduces a Dual Teacher-Student Weight-Sharing Model (DTSWSM) for knowledge distillation from weather-degraded images and a Classifier Weight Updating Attention Mechanism (CWUAM) to dynamically adjust classifier weights based on environmental attributes. The abstract asserts that comprehensive evaluations show significant outperformance over baselines in accuracy and robustness under clear, rainy, cloudy, and foggy conditions.

Significance. If the experimental claims hold with proper validation, the work could contribute to practical all-weather perception systems by combining semi-supervised teacher-student learning with attention-based adaptation, potentially lowering annotation costs for autonomous driving applications. The targeting of real-world weather robustness is a strength. However, the complete absence of quantitative metrics, dataset details, baselines, or ablations in the abstract prevents assessment of whether the result advances the field beyond existing domain-adaptation or robust segmentation methods.

major comments (2)
  1. [Abstract] Abstract: The headline claim that 'WeatherSeg significantly outperforms baseline models in both accuracy and robustness across various weather conditions' is unsupported by any numerical evidence, such as mIoU or pixel accuracy values, dataset names (e.g., adverse-weather variants of Cityscapes or BDD100K), baseline comparisons, or ablation results. This directly undermines evaluation of the central claim.
  2. [Abstract] Abstract: The DTSWSM and CWUAM are introduced at a high level with no loss formulation, no description of the weight-sharing implementation, no specification of how environmental attributes are obtained or injected, and no controls for bias transfer or hyperparameter tuning. These details are load-bearing because the reported outperformance on degraded images depends on successful knowledge transfer without inheriting artifacts and on reliable dynamic reweighting.
minor comments (1)
  1. [Abstract] The abstract uses the phrase 'comprehensive evaluations' without referencing specific tables, figures, or sections that would allow readers to locate the supporting results.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. We agree that the abstract requires strengthening to better support our claims and will revise it accordingly. Below we provide point-by-point responses to the major comments.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The headline claim that 'WeatherSeg significantly outperforms baseline models in both accuracy and robustness across various weather conditions' is unsupported by any numerical evidence, such as mIoU or pixel accuracy values, dataset names (e.g., adverse-weather variants of Cityscapes or BDD100K), baseline comparisons, or ablation results. This directly undermines evaluation of the central claim.

    Authors: We acknowledge that the current abstract does not include specific numerical results or dataset names. The full manuscript reports detailed mIoU and accuracy metrics from experiments on weather-degraded image datasets, with comparisons to multiple baselines and ablation studies. We will revise the abstract to incorporate representative quantitative highlights and dataset references to substantiate the performance claims. revision: yes

  2. Referee: [Abstract] Abstract: The DTSWSM and CWUAM are introduced at a high level with no loss formulation, no description of the weight-sharing implementation, no specification of how environmental attributes are obtained or injected, and no controls for bias transfer or hyperparameter tuning. These details are load-bearing because the reported outperformance on degraded images depends on successful knowledge transfer without inheriting artifacts and on reliable dynamic reweighting.

    Authors: Abstracts conventionally provide high-level overviews, with technical details reserved for the main text. The manuscript details the DTSWSM loss formulations, weight-sharing implementation, environmental attribute injection via attention, and controls for bias transfer and hyperparameters through equations and ablation experiments in Sections 3 and 4. We will partially revise the abstract to briefly reference these mechanisms and direct readers to the detailed descriptions. revision: partial

Circularity Check

0 steps flagged

No derivation chain or equations present; claims rest on empirical evaluations of proposed components

full rationale

The paper introduces WeatherSeg as a semi-supervised framework combining DTSWSM for knowledge distillation and CWUAM for dynamic classifier adjustment, but the abstract and available text contain no equations, loss formulations, or mathematical derivations. Performance claims are supported by evaluations on weather conditions rather than any self-referential reduction, fitted parameters renamed as predictions, or self-citation chains. The method is presented as an empirical solution for all-weather segmentation, self-contained against external benchmarks without load-bearing circular steps.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 2 invented entities

Abstract-only review supplies no explicit free parameters, axioms, or invented physical entities; the two named modules are treated as methodological inventions whose independent evidence is not provided.

invented entities (2)
  • Dual Teacher-Student Weight-Sharing Model (DTSWSM) no independent evidence
    purpose: Enables knowledge distillation from weather-affected images via weight sharing
    Presented as a core component of the framework; no independent validation or external benchmark cited in abstract.
  • Classifier Weight Updating Attention Mechanism (CWUAM) no independent evidence
    purpose: Dynamically adjusts classifier weights according to environmental attributes
    Introduced to handle weather variability; effectiveness asserted without supporting data or prior literature reference in the abstract.

pith-pipeline@v0.9.0 · 5406 in / 1327 out tokens · 54778 ms · 2026-05-10T06:38:33.878379+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

49 extracted references · 49 canonical work pages

  1. [1]

    A brief survey on semantic segmentation with deep learning,

    S. Hao, Y . Zhou, and Y . Guo, “A brief survey on semantic segmentation with deep learning,” Neurocomputing, vol. 406, pp. 302–321, 2020

    work page 2020

  2. [2]

    Review the state-of-the-art technologies of se- mantic segmentation based on deep learning,

    Y . Mo, Y . Wu, X. Yang, F. Liu, and Y . Liao, “Review the state-of-the-art technologies of se- mantic segmentation based on deep learning,” Neurocomputing, vol. 493, pp. 626–646, 2022

    work page 2022

  3. [3]

    Vision-based semantic segmentation in scene understanding for autonomous driving: Recent achievements, challenges, and outlooks,

    K. Muhammad, T. Hussain, H. Ullah, J. D. Ser, M. Rezaei, N. Kumar, M. Hijji, P. Bellavista, and V . H. C. de Albuquerque, “Vision-based semantic segmentation in scene understanding for autonomous driving: Recent achievements, challenges, and outlooks,”IEEE Transactions on Intelligent Transportation Systems, vol. 23, no. 12, pp. 22 694–22 715, 2022

    work page 2022

  4. [4]

    Semantic segmentation for self-driving cars us- ing deep learning: a survey,

    Q. Sellat, S. K. Bisoy, and R. Priyadarshini, “Semantic segmentation for self-driving cars us- ing deep learning: a survey,” inCognitive Big Data Intelligence with a Metaheuristic Approach. Elsevier, 2022, pp. 211–238. 10

    work page 2022

  5. [5]

    Real-time semantic image segmen- tation with deep learning for autonomous driving: A survey,

    I. Papadeas, L. Tsochatzidis, A. Amanatiadis, and I. Pratikakis, “Real-time semantic image segmen- tation with deep learning for autonomous driving: A survey,”Applied Sciences, vol. 11, no. 19, 2021

    work page 2021

  6. [6]

    View-coherent correlation consistency for semi-supervised se- mantic segmentation,

    Y . Hou, S. Gould, and L. Zheng, “View-coherent correlation consistency for semi-supervised se- mantic segmentation,”Pattern Recognition, vol. 147, p. 110089, 2024

    work page 2024

  7. [7]

    Perturbed and strict mean teachers for semi-supervised semantic segmenta- tion,

    Y . Liu, Y . Tian, Y . Chen, F. Liu, V . Belagiannis, and G. Carneiro, “Perturbed and strict mean teachers for semi-supervised semantic segmenta- tion,” inProceedings of the IEEE/CVF Confer- ence on Computer Vision and Pattern Recogni- tion (CVPR), June 2022, pp. 4258–4267

    work page 2022

  8. [8]

    Semi- supervised semantic segmentation using cross- consistency training for pavement crack detec- tion,

    X. Liu, K. Wu, X. Cai, and W. Huang, “Semi- supervised semantic segmentation using cross- consistency training for pavement crack detec- tion,”Road Materials and Pavement Design, vol. 25, no. 6, pp. 1368–1380, 2024

    work page 2024

  9. [9]

    Semi- supervised semantic segmentation with cross- consistency training,

    Y . Ouali, C. Hudelot, and M. Tami, “Semi- supervised semantic segmentation with cross- consistency training,” in2020 IEEE/CVF Con- ference on Computer Vision and Pattern Recog- nition (CVPR), Jun 2020

    work page 2020

  10. [10]

    Semi-supervised semantic segmentation using unreliable pseudo- labels,

    Y . Wang, H. Wang, Y . Shen, J. Fei, W. Li, G. Jin, L. Wu, R. Zhao, and X. Le, “Semi-supervised semantic segmentation using unreliable pseudo- labels,” inProceedings of the IEEE/CVF Confer- ence on Computer Vision and Pattern Recogni- tion (CVPR), June 2022, pp. 4248–4257

    work page 2022

  11. [11]

    Learn- ing pseudo labels for semi-and-weakly super- vised semantic segmentation,

    Y . Wang, J. Zhang, M. Kan, and S. Shan, “Learn- ing pseudo labels for semi-and-weakly super- vised semantic segmentation,”Pattern Recogni- tion, vol. 132, p. 108925, 2022

    work page 2022

  12. [12]

    Semi-supervised semantic segmentation via en- tropy minimization,

    J. Wu, H. Fan, X. Zhang, S. Lin, and Z. Li, “Semi-supervised semantic segmentation via en- tropy minimization,” in2021 IEEE International Conference on Multimedia and Expo (ICME), 2021, pp. 1–6

    work page 2021

  13. [13]

    Semantic segmentation with genera- tive models: Semi-supervised learning and strong out-of-domain generalization,

    D. Li, J. Yang, K. Kreis, A. Torralba, and S. Fidler, “Semantic segmentation with genera- tive models: Semi-supervised learning and strong out-of-domain generalization,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021, pp. 8300–8311

    work page 2021

  14. [14]

    Mean teachers are better role models: Weight-averaged consistency targets improve semi-supervised deep learning re- sults,

    A. Tarvainen and H. Valpola, “Mean teachers are better role models: Weight-averaged consistency targets improve semi-supervised deep learning re- sults,”Advances in neural information processing systems, vol. 30, 2017

    work page 2017

  15. [15]

    Semantic segmentation in autonomous driving—an example of fcn,

    H. Sun and T. Wang, “Semantic segmentation in autonomous driving—an example of fcn,” in Proceedings of the 2023 7th International Con- ference on Innovation in Artificial Intelligence. New York, NY , USA: Association for Computing Machinery, 2023, p. 11–18

    work page 2023

  16. [16]

    Segnet: A deep convolutional encoder-decoder architecture for image segmentation,

    V . Badrinarayanan, A. Kendall, and R. Cipolla, “Segnet: A deep convolutional encoder-decoder architecture for image segmentation,”IEEE transactions on pattern analysis and machine intelligence, vol. 39, no. 12, pp. 2481–2495, 2017

    work page 2017

  17. [17]

    Image segmentation for self-driving car,

    S. Gautam, T. Mathuria, and S. Meena, “Image segmentation for self-driving car,” in2022 2nd International Conference on Intelligent Technolo- gies (CONIT). IEEE, 2022, pp. 1–6

    work page 2022

  18. [18]

    Two-stage framework with improved u-net based on self-supervised contrastive learning for pavement crack segmen- tation,

    Q. Song, W. Yao, H. Tian, Y . Guo, R. C. Muniyandi, and Y . An, “Two-stage framework with improved u-net based on self-supervised contrastive learning for pavement crack segmen- tation,”Expert Systems with Applications, vol. 238, p. 122406, 2024

    work page 2024

  19. [19]

    Deeplab: Semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected crfs,

    L.-C. Chen, G. Papandreou, I. Kokkinos, K. Mur- phy, and A. L. Yuille, “Deeplab: Semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected crfs,”IEEE transactions on pattern analysis and machine intelligence, vol. 40, no. 4, pp. 834–848, 2017

    work page 2017

  20. [20]

    Sats: Self-attention transfer for continual semantic segmentation,

    Y . Qiu, Y . Shen, Z. Sun, Y . Zheng, X. Chang, W. Zheng, and R. Wang, “Sats: Self-attention transfer for continual semantic segmentation,” Pattern Recognition, vol. 138, p. 109383, 2023

    work page 2023

  21. [21]

    Scribblenet: Efficient interactive anno- tation of urban city scenes for semantic segmen- tation,

    B. Sambaturu, A. Gupta, C. Jawahar, and C. Arora, “Scribblenet: Efficient interactive anno- tation of urban city scenes for semantic segmen- tation,”Pattern Recognition, vol. 133, p. 109011, 2023

    work page 2023

  22. [22]

    Simpler is better: Few-shot semantic segmentation with classifier weight transformer,

    Z. Lu, S. He, X. Zhu, L. Zhang, Y .-Z. Song, and T. Xiang, “Simpler is better: Few-shot semantic segmentation with classifier weight transformer,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2021, pp. 8741– 8750

    work page 2021

  23. [23]

    Beyond low- dimensional features: Enhancing semi-supervised medical image semantic segmentation with ad- vanced consistency learning techniques,

    Y . Lu, W. Li, Z. Cui, and Y . Zhang, “Beyond low- dimensional features: Enhancing semi-supervised medical image semantic segmentation with ad- vanced consistency learning techniques,”Expert Systems with Applications, p. 125456, 2024

    work page 2024

  24. [24]

    Semi-supervised semantic segmentation with prototype-based consistency regularization,

    H. Xu, L. Liu, Q. Bian, and Z. Yang, “Semi-supervised semantic segmentation with prototype-based consistency regularization,”Ad- vances in Neural Information Processing Sys- tems, vol. 35, pp. 26 007–26 020, 2022

    work page 2022

  25. [25]

    Enhancing semi-supervised semantic segmen- tation of remote sensing images via fea- ture perturbation-based consistency regulariza- tion methods,

    Y . Xin, Z. Fan, X. Qi, Y . Geng, and X. Li, “Enhancing semi-supervised semantic segmen- tation of remote sensing images via fea- ture perturbation-based consistency regulariza- tion methods,”Sensors, vol. 24, no. 3, p. 730, 2024

    work page 2024

  26. [26]

    Improving semi-supervised and domain-adaptive semantic segmentation with self-supervised depth estimation,

    L. Hoyer, D. Dai, Q. Wang, Y . Chen, and L. Van Gool, “Improving semi-supervised and domain-adaptive semantic segmentation with self-supervised depth estimation,”International Journal of Computer Vision, pp. 1–27, 2023

    work page 2023

  27. [27]

    Learning from pixel-level label noise: A new perspective for semi-supervised semantic seg- mentation,

    R. Yi, Y . Huang, Q. Guan, M. Pu, and R. Zhang, 11 “Learning from pixel-level label noise: A new perspective for semi-supervised semantic seg- mentation,”IEEE Transactions on Image Pro- cessing, vol. 31, pp. 623–635, 2022

    work page 2022

  28. [28]

    Semi-supervised remote sensing image semantic segmentation via consistency regular- ization and average update of pseudo-label,

    J. Wang, C. HQ Ding, S. Chen, C. He, and B. Luo, “Semi-supervised remote sensing image semantic segmentation via consistency regular- ization and average update of pseudo-label,”Re- mote Sensing, vol. 12, no. 21, p. 3603, 2020

    work page 2020

  29. [29]

    Ambiguity-selective con- sistency regularization for mean-teacher semi- supervised medical image segmentation,

    Z. Xu, Y . Wang, D. Lu, X. Luo, J. Yan, Y . Zheng, and R. K.-y. Tong, “Ambiguity-selective con- sistency regularization for mean-teacher semi- supervised medical image segmentation,”Med- ical Image Analysis, vol. 88, p. 102880, 2023

    work page 2023

  30. [30]

    Dual attention based uncertainty-aware mean teacher model for semi-supervised cardiac image segmentation,

    A. Xu, S. Wang, J. Fan, X. Shi, and Q. Chen, “Dual attention based uncertainty-aware mean teacher model for semi-supervised cardiac image segmentation,” in2021 IEEE International Con- ference on Progress in Informatics and Comput- ing (PIC). IEEE, 2021, pp. 82–86

    work page 2021

  31. [31]

    Semi-supervised brain lesion segmentation with an adapted mean teacher model,

    W. Cui, Y . Liu, Y . Li, M. Guo, Y . Li, X. Li, T. Wang, X. Zeng, and C. Ye, “Semi-supervised brain lesion segmentation with an adapted mean teacher model,” inInformation Processing in Medical Imaging: 26th International Conference, IPMI 2019, Hong Kong, China, June 2–7, 2019, Proceedings 26. Springer, 2019, pp. 554–565

    work page 2019

  32. [32]

    Semi-supervised semantic segmen- tation with cross teacher training,

    H. Xiao, D. Li, H. Xu, S. Fu, D. Yan, K. Song, and C. Peng, “Semi-supervised semantic segmen- tation with cross teacher training,”Neurocomput- ing, vol. 508, pp. 36–46, 2022

    work page 2022

  33. [33]

    Semi-supervised semantic segmentation via gentle teaching assis- tant,

    Y . Jin, J. Wang, and D. Lin, “Semi-supervised semantic segmentation via gentle teaching assis- tant,”Advances in Neural Information Processing Systems, vol. 35, pp. 2803–2816, 2022

    work page 2022

  34. [34]

    Automated evaluation of semantic segmentation robustness for autonomous driving,

    W. Zhou, J. S. Berrio, S. Worrall, and E. Nebot, “Automated evaluation of semantic segmentation robustness for autonomous driving,”IEEE Trans- actions on Intelligent Transportation Systems, vol. 21, no. 5, pp. 1951–1963, 2020

    work page 1951

  35. [35]

    Deep cluster- ing for weakly-supervised semantic segmentation in autonomous driving scenes,

    X. Wang, H. Ma, and S. You, “Deep cluster- ing for weakly-supervised semantic segmentation in autonomous driving scenes,”Neurocomputing, vol. 381, pp. 20–28, 2020

    work page 2020

  36. [36]

    Acdc: The adverse conditions dataset with correspon- dences for semantic driving scene understand- ing,

    C. Sakaridis, D. Dai, and L. Van Gool, “Acdc: The adverse conditions dataset with correspon- dences for semantic driving scene understand- ing,” inProceedings of the IEEE/CVF Interna- tional Conference on Computer Vision (ICCV), October 2021, pp. 10 765–10 775

    work page 2021

  37. [37]

    Rainy wcity: A real rainfall dataset with diverse conditions for semantic driv- ing scene understanding

    X. Zhong, S. Tu, X. Ma, K. Jiang, W. Huang, and Z. Wang, “Rainy wcity: A real rainfall dataset with diverse conditions for semantic driv- ing scene understanding.” inIJCAI, 2022, pp. 1743–1749

    work page 2022

  38. [38]

    Revisiting weak-to-strong consistency in semi- supervised semantic segmentation,

    L. Yang, L. Qi, L. Feng, W. Zhang, and Y . Shi, “Revisiting weak-to-strong consistency in semi- supervised semantic segmentation,” inProceed- ings of the IEEE/CVF conference on computer vision and pattern recognition, 2023, pp. 7236– 7246

    work page 2023

  39. [39]

    Semi-supervised semantic segmen- tation via adaptive equalization learning,

    H. Hu, F. Wei, H. Hu, Q. Ye, J. Cui, and L. Wang, “Semi-supervised semantic segmen- tation via adaptive equalization learning,”Ad- vances in Neural Information Processing Sys- tems, vol. 34, pp. 22 106–22 118, 2021

    work page 2021

  40. [40]

    Semi-supervised semantic segmentation with cross pseudo supervision,

    X. Chen, Y . Yuan, G. Zeng, and J. Wang, “Semi-supervised semantic segmentation with cross pseudo supervision,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), June 2021, pp. 2613–2622

    work page 2021

  41. [41]

    En- hanced soft label for semi-supervised semantic segmentation,

    J. Ma, C. Wang, Y . Liu, L. Lin, and G. Li, “En- hanced soft label for semi-supervised semantic segmentation,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 1185–1195

    work page 2023

  42. [42]

    The cityscapes dataset for se- mantic urban scene understanding,

    M. Cordts, M. Omran, S. Ramos, T. Rehfeld, M. Enzweiler, R. Benenson, U. Franke, S. Roth, and B. Schiele, “The cityscapes dataset for se- mantic urban scene understanding,” in2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Jun 2016

    work page 2016

  43. [43]

    The pascal vi- sual object classes (voc) challenge,

    M. Everingham, L. Van Gool, C. K. I. Williams, J. Winn, and A. Zisserman, “The pascal vi- sual object classes (voc) challenge,”International Journal of Computer Vision, p. 303–338, Jun 2010

    work page 2010

  44. [44]

    Z. Ke, D. Qiu, K. Li, Q. Yan, and R. W. H. Lau,Guided Collaborative Training for Pixel- wise Semi-Supervised Learning, Jan 2020, p. 429–445

    work page 2020

  45. [45]

    Semi-supervised seman- tic segmentation needs strong, high-dimensional perturbations,

    G. French, T. Aila, S. Laine, M. Mackiewicz, and G. D. Finlayson, “Semi-supervised seman- tic segmentation needs strong, high-dimensional perturbations,” Sep 2019

    work page 2019

  46. [46]

    Semi-supervised semantic segmentation using unreliable pseudo- labels,

    Y . Wang, H. Wang, Y . Shen, J. Fei, W. Li, G. Jin, L. Wu, R. Zhao, and X. Le, “Semi-supervised semantic segmentation using unreliable pseudo- labels,” inProceedings of the IEEE/CVF confer- ence on computer vision and pattern recognition, 2022, pp. 4248–4257

    work page 2022

  47. [47]

    Semi-supervised seman- tic segmentation with error localization network,

    D. Kwon and S. Kwak, “Semi-supervised seman- tic segmentation with error localization network,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 9957–9967

    work page 2022

  48. [48]

    Augmentation matters: A simple-yet- effective approach to semi-supervised semantic segmentation,

    Z. Zhao, L. Yang, S. Long, J. Pi, L. Zhou, and J. Wang, “Augmentation matters: A simple-yet- effective approach to semi-supervised semantic segmentation,” inProceedings of the IEEE/CVF conference on computer vision and pattern recog- nition, 2023, pp. 11 350–11 359. 12 Dr. Zhang Zhangis currently a lecturer at the School of Artificial Intelligence and Inf...

    work page 2023

  49. [49]

    His research interests include intelli- gent agents, decision making, social net- works, and computer games

    He is a Professor and Head of Research and Knowledge Exchange with the Department of Computer & Infor- mation Sciences, Northumbria University, UK. His research interests include intelli- gent agents, decision making, social net- works, and computer games. Most of his publications appear in the most prestigious international academic journals and conferen...

    work page 2019