pith. sign in

arxiv: 2605.18156 · v1 · pith:VNN3464Anew · submitted 2026-05-18 · 💻 cs.CV

Semi-LAR: Semi-supervised Contrastive Learning with Linear Attention for Removal of Nighttime Flares

Xiyu Zhu , Wei Wang , Kui Jiang , Zhengguo Li This is my paper

Pith reviewed 2026-05-20 11:38 UTC · model grok-4.3

classification 💻 cs.CV
keywords semi-supervised learninglens flare removalcontrastive learningnighttime flarespseudo-label refinementimage restorationlinear attention
0
0 comments X

The pith

A semi-supervised framework removes nighttime lens flares by refining pseudo-labels with quality assessment and applying flare-aware contrastive learning to unlabeled images.

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

The paper presents Semi-LAR, a semi-supervised method for removing lens flares in nighttime images that reduces dependence on large paired datasets. It maintains an adaptive pseudo-label repository that evaluates and updates labels using no-reference quality scores, momentum updates, and filtering to avoid error buildup during training. A dedicated contrastive loss then treats flare-affected inputs as negatives and operates at the patch level to build representations that separate flares from scene content. This joint handling of label reliability and discrimination allows the model to learn stably from unlabeled data and improves results across benchmarks in a way that works with different base models.

Core claim

The authors claim that jointly managing pseudo-label reliability via an adaptive repository (with no-reference quality assessment, momentum-based updates, and invalid label filtering) and representation discrimination via a flare-aware contrastive loss (treating flare-contaminated inputs as negatives for patch-level learning) enables stable semi-supervised training for nighttime flare removal, yielding consistent gains in performance and robustness on multiple benchmarks while remaining model-agnostic.

What carries the argument

The adaptive pseudo-label repository paired with the flare-aware contrastive loss, where the repository progressively refines pseudo supervision and the loss encourages discriminative features against flare patterns while aligning with reliable targets.

If this is right

  • Existing flare removal networks can be wrapped with this framework to gain performance from unlabeled nighttime images.
  • Training stability increases because poor pseudo-labels are detected and filtered before they compound.
  • The learned features become more robust to flare patterns while staying faithful to the underlying scene.
  • The approach works across different backbone architectures without requiring architecture-specific changes.

Where Pith is reading between the lines

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

  • The same quality-guided pseudo-label refinement could transfer to other low-light restoration tasks where paired data is expensive to create.
  • If linear attention is used in the underlying network as the title suggests, the method may support faster inference on edge devices for real-time flare removal.
  • Future work could test whether the contrastive component alone suffices when high-quality no-reference metrics are unavailable.

Load-bearing premise

No-reference quality assessment can reliably evaluate and progressively refine pseudo-labels to prevent error accumulation while the flare-aware contrastive loss produces discriminative representations without degrading scene fidelity.

What would settle it

Running the framework on a standard flare removal benchmark and finding no gain or a drop in standard metrics such as PSNR or visual quality relative to a fully supervised baseline trained on the same labeled data would falsify the claim.

Figures

Figures reproduced from arXiv: 2605.18156 by Kui Jiang, Wei Wang, Xiyu Zhu, Zhengguo Li.

Figure 1
Figure 1. Figure 1: Our approach bridges the gap to full supervision under limited labels and achieves further quality improvements when added to existing labeled baselines. ity but can also disrupt downstream vision tasks that rely on stable edges, consistent illumination, and accurate color statistics (Li et al., 2025b; Girshick et al., 2014). Similar nighttime visibility degradation has also been studied from the perspecti… view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the proposed semi-supervised flare removal framework. The method adopts a teacher–student paradigm with EMA-based teacher updating. Labeled data are optimized with supervised reconstruction losses, while unlabeled data are processed using weak and strong augmentations. Teacher predictions are evaluated and selectively stored in a dependable repository to provide reliable pseudo supervision. The… view at source ↗
Figure 3
Figure 3. Figure 3: Overview of the proposed RaLiFormer architecture. (a) The hierarchical U-shaped network structure for image restoration. (b) The Directionally-Enhanced Spatial Module (DESM). (c) The Parallel Spatial-Channel Block (PSCBlock) as the basic building unit. (d) The Rank-Enhanced Linear Attention (ReLinA) module. (e) Detailed structure of the Channel Attention Block module. output yˆu = fθ(x s u ), while a relia… view at source ↗
Figure 4
Figure 4. Figure 4: Visual comparison of flare removal on real-world nighttime flare images. * indicates models trained with 8K labeled and 8K unlabeled data. Linear Attention (ReLinA), which enables efficient global context modeling with linear complexity. Given query, key, and value projections (Q, K, V), a positive feature mapping is applied: Q′ = ϕ(Q), K′ = ϕ(K), ϕ(x) = 1 + ELU(x). (6) The attention output is computed as:… view at source ↗
Figure 5
Figure 5. Figure 5: Ablation studies on the proposed method. W/O All W/O RG W/O CJ W/O GB Ours [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Ablation studies on the impact of strong augmentations . learning methods trained on the full Flare7K++ dataset, such as U-Net (Ronneberger et al., 2015), Restormer (Zamir et al., 2022), Uformer (Wang et al., 2022), DeflareMamba (Huang et al., 2025), and SGSFT (Ma et al., 2025), which serve as strong upper-bound baselines. (3) Limited-supervision and semi-supervised methods, including limited-data variants… view at source ↗
Figure 7
Figure 7. Figure 7: Visual comparison of flare removal on flare-corrupted test. * indicates models trained with 8K labeled and 8K unlabeled data. 15 [PITH_FULL_IMAGE:figures/full_fig_p015_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Visual comparison of flare removal on FlareX test. * indicates models trained with 8K labeled and 8K unlabeled data. 16 [PITH_FULL_IMAGE:figures/full_fig_p016_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Visual comparison of flare removal on FlareReal600 test. * indicates models trained with 8K labeled and 8K unlabeled data. 17 [PITH_FULL_IMAGE:figures/full_fig_p017_9.png] view at source ↗
read the original abstract

Lens flare removal is challenging due to the large spatial extent of flare artifacts and their entanglement with scene structures, while existing methods heavily rely on large-scale paired data. We propose a semi-supervised flare removal framework that enables stable learning from unlabeled images by jointly addressing pseudo-label reliability and representation discrimination. We propose an adaptive pseudo-label repository that progressively refines pseudo supervision through no-reference quality assessment, momentum-based updates, and invalid label filtering, effectively mitigating error accumulation. Moreover, we propose a flare-aware contrastive loss that explicitly treats flare-contaminated inputs as negatives and performs patch-level contrastive learning, encouraging representations that are discriminative against flare patterns while remaining consistent with reliable pseudo targets. Extensive experiments on multiple flare benchmarks demonstrate that the proposed framework is model-agnostic and consistently improves performance and robustness.

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 / 2 minor

Summary. The paper claims to introduce Semi-LAR, a semi-supervised framework for nighttime lens flare removal that reduces reliance on paired data. It proposes an adaptive pseudo-label repository that progressively refines supervision from unlabeled images using no-reference quality assessment, momentum-based updates, and invalid label filtering to mitigate error accumulation. It further introduces a flare-aware contrastive loss that treats flare-contaminated inputs as negatives and performs patch-level contrastive learning to produce representations discriminative against flare patterns while remaining consistent with reliable pseudo targets. The framework is described as model-agnostic, with extensive experiments on multiple flare benchmarks claimed to show consistent improvements in performance and robustness.

Significance. If the central claims hold, the work could meaningfully advance semi-supervised approaches to image restoration tasks involving complex, spatially extended artifacts like nighttime flares. The joint focus on pseudo-label reliability and flare-specific representation discrimination via contrastive learning offers a practical way to leverage unlabeled data, and the model-agnostic design is a positive feature. Credit is given for attempting to create a stable learning pipeline without paired supervision, though the overall impact hinges on whether the no-reference IQA component reliably proxies true flare-removal quality.

major comments (2)
  1. [Abstract] Abstract: The central claim that the adaptive pseudo-label repository enables stable learning from unlabeled images by mitigating error accumulation rests on the assumption that no-reference quality assessment reliably ranks pseudo-labels by actual flare-removal fidelity. In nighttime scenes, where flares are spatially extended and entangled with scene content, this proxy may not correlate with ground-truth quality metrics; if misaligned, momentum updates would reinforce incorrect labels and the subsequent flare-aware contrastive loss would optimize for the wrong discrimination boundary. A correlation analysis or ablation comparing the chosen no-reference metric(s) against paired-data metrics (e.g., PSNR) on a validation subset is required to substantiate this load-bearing component.
  2. [Method] Method section describing the flare-aware contrastive loss: The claim that patch-level contrastive learning with flare-contaminated inputs as negatives produces discriminative representations without degrading scene fidelity lacks supporting detail on negative sampling strategy, temperature scaling, or how consistency with pseudo targets is enforced. Without these specifics or an ablation isolating the loss contribution, it is unclear whether the loss actually improves robustness or merely trades one form of artifact for another.
minor comments (2)
  1. [Abstract] The abstract states that the framework is 'model-agnostic' but provides no explicit description of which backbone architectures were tested or how the linear attention module integrates with them.
  2. [Abstract] No implementation details (e.g., specific no-reference IQA metrics employed, momentum coefficient values, or filtering thresholds) are mentioned in the abstract, which hinders immediate reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments, which help clarify key assumptions and implementation details in our semi-supervised framework. We address each major comment below and will revise the manuscript accordingly.

read point-by-point responses

  1. Referee: [Abstract] The central claim that the adaptive pseudo-label repository enables stable learning from unlabeled images by mitigating error accumulation rests on the assumption that no-reference quality assessment reliably ranks pseudo-labels by actual flare-removal fidelity. In nighttime scenes, where flares are spatially extended and entangled with scene content, this proxy may not correlate with ground-truth quality metrics; if misaligned, momentum updates would reinforce incorrect labels and the subsequent flare-aware contrastive loss would optimize for the wrong discrimination boundary. A correlation analysis or ablation comparing the chosen no-reference metric(s) against paired-data metrics (e.g., PSNR) on a validation subset is required to substantiate this load-bearing component.

    Authors: We agree that the correlation between no-reference quality assessment and true flare-removal fidelity (measured by paired metrics such as PSNR) is a critical assumption requiring explicit validation. In the revised manuscript, we will add a dedicated ablation study and correlation analysis performed on a held-out validation subset containing paired ground-truth data. This will quantify how well the chosen no-reference metric ranks pseudo-labels and will directly address concerns about potential misalignment or error reinforcement in the momentum updates. revision: yes

  2. Referee: [Method] The claim that patch-level contrastive learning with flare-contaminated inputs as negatives produces discriminative representations without degrading scene fidelity lacks supporting detail on negative sampling strategy, temperature scaling, or how consistency with pseudo targets is enforced. Without these specifics or an ablation isolating the loss contribution, it is unclear whether the loss actually improves robustness or merely trades one form of artifact for another.

    Authors: We acknowledge that the current description of the flare-aware contrastive loss requires additional implementation specifics for clarity and reproducibility. In the revised method section, we will expand the exposition to detail the negative sampling strategy (including how flare-contaminated patches are identified and selected as negatives), the temperature scaling hyperparameter, and the explicit consistency enforcement mechanism with reliable pseudo targets. We will also include a new ablation isolating the contribution of this loss to demonstrate its effect on robustness without introducing alternative artifacts. revision: yes

Circularity Check

0 steps flagged

No circularity: descriptive method proposal without self-referential reductions

full rationale

The provided abstract and manuscript description present a semi-supervised flare removal framework through natural-language claims about an adaptive pseudo-label repository (no-reference quality assessment, momentum updates, invalid label filtering) and a flare-aware contrastive loss. No equations, fitted parameters, or derivation chains appear that would reduce any prediction or result to quantities defined by the method itself. No self-citations are invoked as load-bearing uniqueness theorems or ansatzes. The central claims remain independent contributions rather than tautological restatements of inputs, making this a standard non-circular finding for a descriptive algorithmic proposal.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract provides no explicit free parameters, mathematical axioms, or newly postulated entities; all components are described at the level of algorithmic choices rather than formal assumptions.

pith-pipeline@v0.9.0 · 5668 in / 1092 out tokens · 42661 ms · 2026-05-20T11:38:56.811543+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

114 extracted references · 114 canonical work pages

  1. [1]

    Chen, G.-Y., Dong, W., Fan, G., Su, J.-N., Gan, M., and Chen, C. P. Lpfsformer: Location prior guided frequency and spatial interactive learning for nighttime flare removal. IEEE Transactions on Circuits and Systems for Video Technology, 2024

    work page 2024

  2. [2]

    A simple framework for contrastive learning of visual representations

    Chen, T., Kornblith, S., Norouzi, M., and Hinton, G. A simple framework for contrastive learning of visual representations. Cornell University - arXiv,Cornell University - arXiv, Feb 2020

    work page 2020

  3. [3]

    A semi-supervised nighttime dehazing baseline with spatial-frequency aware and realistic brightness constraint

    Cong, X., Gui, J., Zhang, J., Hou, J., and Shen, H. A semi-supervised nighttime dehazing baseline with spatial-frequency aware and realistic brightness constraint. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp.\ 2631--2640, 2024

    work page 2024

  4. [4]

    Dai, Y., Li, C., Zhou, S., Feng, R., and Loy, C. C. Flare7k: A phenomenological nighttime flare removal dataset. In Thirty-sixth Conference on Neural Information Processing Systems Datasets and Benchmarks Track, 2022

    work page 2022

  5. [5]

    Dai, Y., Li, C., Zhou, S., Feng, R., Luo, Y., and Loy, C. C. Flare7k++: Mixing synthetic and real datasets for nighttime flare removal and beyond. 2023 a

    work page 2023

  6. [6]

    C., Gu, J., Liu, S., et al

    Dai, Y., Li, C., Zhou, S., Feng, R., Zhu, Q., Sun, Q., Sun, W., Loy, C. C., Gu, J., Liu, S., et al. Mipi 2023 challenge on nighttime flare removal: Methods and results. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp.\ 2853--2863, 2023 b

    work page 2023

  7. [7]

    Ernst, M., Akenine-M \"o ller, T., and Jensen, H. W. Interactive rendering of caustics using interpolated warped volumes. In Proceedings of Graphics Interface 2005, pp.\ 87--96, 2005

    work page 2005

  8. [8]

    Rich feature hierarchies for accurate object detection and semantic segmentation

    Girshick, R., Donahue, J., Darrell, T., and Malik, J. Rich feature hierarchies for accurate object detection and semantic segmentation. In Proceedings of the IEEE conference on computer vision and pattern recognition, pp.\ 580--587, 2014

    work page 2014

  9. [9]

    Disentangle nighttime lens flares: self-supervised generation-based lens flare removal

    He, Y., Wang, W., Wu, W., and Jiang, K. Disentangle nighttime lens flares: self-supervised generation-based lens flare removal. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 39, pp.\ 3464--3472, 2025

    work page 2025

  10. [10]

    Deflaremamba: Hierarchical vision mamba for contextually consistent lens flare removal

    Huang, Y., Huang, Y., Lin, J., and Huang, H. Deflaremamba: Hierarchical vision mamba for contextually consistent lens flare removal. In Proceedings of the 33rd ACM International Conference on Multimedia, pp.\ 8028--8037, 2025

    work page 2025

  11. [11]

    Physically-based real-time lens flare rendering

    Hullin, M., Eisemann, E., Seidel, H.-P., and Lee, S. Physically-based real-time lens flare rendering. In ACM SIGGRAPH 2011 papers, pp.\ 1--10. 2011

    work page 2011

  12. [12]

    Mfdnet: Multi-frequency deflare network for efficient nighttime flare removal

    Jiang, Y., Chen, X., Pun, C.-M., Wang, S., and Feng, W. Mfdnet: Multi-frequency deflare network for efficient nighttime flare removal. The Visual Computer, 40 0 (11): 0 7575--7588, 2024. doi:10.1007/s00371-024-03540-x

    work page doi:10.1007/s00371-024-03540-x 2024

  13. [13]

    Jin, Y., Lin, B., Yan, W., Yuan, Y., Ye, W., and Tan, R. T. Enhancing visibility in nighttime haze images using guided apsf and gradient adaptive convolution. In Proceedings of the 31st ACM international conference on multimedia, pp.\ 2446--2457, 2023 a

    work page 2023

  14. [15]

    Musiq: Multi-scale image quality transformer

    Ke, J., Wang, Q., Wang, Y., Milanfar, P., and Yang, F. Musiq: Multi-scale image quality transformer. In Proceedings of the IEEE/CVF international conference on computer vision, pp.\ 5148--5157, 2021

    work page 2021

  15. [16]

    and Torki, M

    Kotp, Y. and Torki, M. Flare-free vision: Empowering uformer with depth insights. In ICASSP 2024-2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp.\ 2565--2569. IEEE, 2024

    work page 2024

  16. [17]

    Crafting papers on machine learning

    Langley, P. Crafting papers on machine learning. In Langley, P. (ed.), Proceedings of the 17th International Conference on Machine Learning (ICML 2000), pp.\ 1207--1216, Stanford, CA, 2000. Morgan Kaufmann

    work page 2000

  17. [18]

    Lee, D.-H. et al. Pseudo-label: The simple and efficient semi-supervised learning method for deep neural networks. In Workshop on challenges in representation learning, ICML, volume 3, pp.\ 896. Atlanta, 2013

    work page 2013

  18. [19]

    Semi-llie: Semi-supervised contrastive learning with mamba-based low-light image enhancement

    Li, G., Zhang, K., Wang, T., Li, M., Zhao, B., and Li, X. Semi-llie: Semi-supervised contrastive learning with mamba-based low-light image enhancement. Neural Networks, pp.\ 108010, 2025 a

    work page 2025

  19. [20]

    Blind 3d video stabilization with spatio-temporally varying motion blur

    Li, H., Wang, W., Wang, X., Yuan, X., and Xu, X. Blind 3d video stabilization with spatio-temporally varying motion blur. ACM Transactions on Multimedia Computing, Communications and Applications, 20 0 (11): 0 1--23, 2024

    work page 2024

  20. [21]

    Neural-augmented hdr imaging via two aligned large-exposure-ratio images

    Li, Z., Zheng, C., Chen, B., and Wu, S. Neural-augmented hdr imaging via two aligned large-exposure-ratio images. IEEE Transactions on Instrumentation and Measurement, 2025 b

    work page 2025

  21. [22]

    Smfr-net: simple multi-domain flare removal network

    Liu, S., Chen, G., Zhang, W., Zhang, Q., Zhang, J., and Wang, J. Smfr-net: simple multi-domain flare removal network. Scientific Reports, 15 0 (1): 0 37251, 2025

    work page 2025

  22. [23]

    Luformer: A luminance-informed localized transformer with frequency augmentation for nighttime flare removal

    Lu, W., Zhao, H., Ma, D., and Jing, P. Luformer: A luminance-informed localized transformer with frequency augmentation for nighttime flare removal. Neural Networks, pp.\ 107660, 2025

    work page 2025

  23. [24]

    Self-prior guided spatial and fourier transformer for nighttime flare removal

    Ma, T., Kai, Z., Miao, X., Liang, J., Peng, J., Wang, Y., Wang, H., and Liu, X. Self-prior guided spatial and fourier transformer for nighttime flare removal. IEEE Transactions on Automation Science and Engineering, 2025

    work page 2025

  24. [25]

    Fcnet: A feature complementary network for nighttime flare removal

    Qi, K., Wang, B., and Li, C. Fcnet: A feature complementary network for nighttime flare removal. Computer Vision and Image Understanding, pp.\ 104495, 2025 a

    work page 2025

  25. [26]

    A self-prompt based dual-domain network for nighttime flare removal

    Qi, K., Wang, B., and Liu, Y. A self-prompt based dual-domain network for nighttime flare removal. Engineering Applications of Artificial Intelligence, 144: 0 110103, 2025 b

    work page 2025

  26. [27]

    Wduie: wavelet-based diffusion for semi-supervised underwater image enhancement

    Qian, F., Ma, L., Su, K., Zhang, Z., Chen, Z., and Zhang, J. Wduie: wavelet-based diffusion for semi-supervised underwater image enhancement. Optics & Laser Technology, 192: 0 113988, 2025

    work page 2025

  27. [28]

    P., and Lau, R

    Qiao, X., Hancke, G. P., and Lau, R. W. Light source guided single-image flare removal from unpaired data. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pp.\ 4177--4185, 2021

    work page 2021

  28. [30]

    K., Ganesh, V

    Raut, H. K., Ganesh, V. A., Nair, A. S., and Ramakrishna, S. Anti-reflective coatings: A critical, in-depth review. Energy & Environmental Science, 4 0 (10): 0 3779--3804, 2011

    work page 2011

  29. [31]

    U-net: Convolutional networks for biomedical image segmentation

    Ronneberger, O., Fischer, P., and Brox, T. U-net: Convolutional networks for biomedical image segmentation. In International Conference on Medical image computing and computer-assisted intervention, pp.\ 234--241. Springer, 2015

    work page 2015

  30. [32]

    and Valpola, H

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

    work page 2017

  31. [33]

    and Ballester, C

    Vitoria, P. and Ballester, C. Automatic flare spot artifact detection and removal in photographs. Journal of Mathematical Imaging and Vision, 61 0 (4): 0 515--533, 2019

    work page 2019

  32. [34]

    and Xu, X

    Wang, W. and Xu, X. Lightweight cnn-based low-light-image enhancement system on fpga platform. Neural Processing Letters, 55 0 (6): 0 8023--8039, 2023

    work page 2023

  33. [35]

    C., Sheikh, H

    Wang, Z., Bovik, A. C., Sheikh, H. R., and Simoncelli, E. P. Image quality assessment: from error visibility to structural similarity. IEEE transactions on image processing, 13 0 (4): 0 600--612, 2004

    work page 2004

  34. [36]

    Uformer: A general u-shaped transformer for image restoration

    Wang, Z., Cun, X., Bao, J., Zhou, W., Liu, J., and Li, H. Uformer: A general u-shaped transformer for image restoration. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp.\ 17683--17693, 2022

    work page 2022

  35. [38]

    Flare removal model based on sparse-uformer networks

    Wu, S., Liu, F., Bai, Y., Han, H., Wang, J., and Zhang, N. Flare removal model based on sparse-uformer networks. Entropy, 26 0 (8): 0 627, 2024 a

    work page 2024

  36. [39]

    For overall nighttime visibility: Integrate irregular glow removal with glow-aware enhancement

    Wu, W., Wang, W., Wang, Z., Jiang, K., and Li, Z. For overall nighttime visibility: Integrate irregular glow removal with glow-aware enhancement. IEEE Transactions on Circuits and Systems for Video Technology, 35 0 (1): 0 823--837, 2024 b

    work page 2024

  37. [40]

    Wu, Y., He, Q., Xue, T., Garg, R., Chen, J., Veeraraghavan, A., and Barron, J. T. How to train neural networks for flare removal. In Proceedings of the IEEE/CVF international conference on computer vision, pp.\ 2239--2247, 2021 b

    work page 2021

  38. [41]

    Xie, Q., Luong, M.-T., Hovy, E., and Le, Q. V. Self-training with noisy student improves imagenet classification. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp.\ 10687--10698, 2020

    work page 2020

  39. [42]

    and Tu, Z

    Xie, S. and Tu, Z. Holistically-nested edge detection. In Proceedings of the IEEE international conference on computer vision, pp.\ 1395--1403, 2015

    work page 2015

  40. [44]

    W., Arora, A., Khan, S., Hayat, M., Khan, F

    Zamir, S. W., Arora, A., Khan, S., Hayat, M., Khan, F. S., and Yang, M.-H. Restormer: Efficient transformer for high-resolution image restoration. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp.\ 5728--5739, 2022

    work page 2022

  41. [45]

    Ff-former: Swin fourier transformer for nighttime flare removal

    Zhang, D., Ouyang, J., Liu, G., Wang, X., Kong, X., and Jin, Z. Ff-former: Swin fourier transformer for nighttime flare removal. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp.\ 2824--2832, 2023

    work page 2023

  42. [46]

    A., Shechtman, E., and Wang, O

    Zhang, R., Isola, P., Efros, A. A., Shechtman, E., and Wang, O. The unreasonable effectiveness of deep features as a perceptual metric. In Proceedings of the IEEE conference on computer vision and pattern recognition, pp.\ 586--595, 2018 a

    work page 2018

  43. [47]

    Flare-aware rwkv for flare removal

    Zhang, W., Shang, W., Ren, D., and Zuo, W. Flare-aware rwkv for flare removal. In ICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp.\ 1--5. IEEE, 2025

    work page 2025

  44. [48]

    Single image reflection separation with perceptual losses

    Zhang, X., Ng, R., and Chen, Q. Single image reflection separation with perceptual losses. In Proceedings of the IEEE conference on computer vision and pattern recognition, pp.\ 4786--4794, 2018 b

    work page 2018

  45. [49]

    Improving lens flare removal with general-purpose pipeline and multiple light sources recovery

    Zhou, Y., Liang, D., Chen, S., Huang, S.-J., Yang, S., and Li, C. Improving lens flare removal with general-purpose pipeline and multiple light sources recovery. In Proceedings of the IEEE/CVF international conference on computer vision, pp.\ 12969--12979, 2023

    work page 2023

  46. [50]

    and Lee, S

    Zhu, J. and Lee, S. Pbfg: A new physically-based dataset and removal of lens flares and glares. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pp.\ 5448--5457, October 2025

    work page 2025

  47. [52]

    Langley , title =

    P. Langley , title =. Proceedings of the 17th International Conference on Machine Learning (ICML 2000) , address =. 2000 , pages =

    work page 2000

  48. [53]

    T. M. Mitchell. The Need for Biases in Learning Generalizations. 1980

    work page 1980

  49. [54]

    M. J. Kearns , title =

    work page

  50. [55]

    Machine Learning: An Artificial Intelligence Approach, Vol. I. 1983

    work page 1983

  51. [56]

    R. O. Duda and P. E. Hart and D. G. Stork. Pattern Classification. 2000

    work page 2000

  52. [57]

    Suppressed for Anonymity , author=

    work page

  53. [58]

    Newell and P

    A. Newell and P. S. Rosenbloom. Mechanisms of Skill Acquisition and the Law of Practice. Cognitive Skills and Their Acquisition. 1981

    work page 1981

  54. [59]

    A. L. Samuel. Some Studies in Machine Learning Using the Game of Checkers. IBM Journal of Research and Development. 1959

    work page 1959

  55. [60]

    Proceedings of the IEEE/CVF international conference on computer vision , pages=

    How to train neural networks for flare removal , author=. Proceedings of the IEEE/CVF international conference on computer vision , pages=

    work page

  56. [61]

    Thirty-sixth Conference on Neural Information Processing Systems Datasets and Benchmarks Track , year=

    Flare7K: A Phenomenological Nighttime Flare Removal Dataset , author=. Thirty-sixth Conference on Neural Information Processing Systems Datasets and Benchmarks Track , year=

    work page

  57. [62]

    Flare7K++: Mixing Synthetic and Real Datasets for Nighttime Flare Removal and Beyond , author=

    work page

  58. [63]

    arXiv preprint arXiv:2306.15884 , year=

    Toward real flare removal: A comprehensive pipeline and a new benchmark , author=. arXiv preprint arXiv:2306.15884 , year=

    work page arXiv

  59. [64]

    arXiv preprint arXiv:2510.09995 , year=

    FlareX: A Physics-Informed Dataset for Lens Flare Removal via 2D Synthesis and 3D Rendering , author=. arXiv preprint arXiv:2510.09995 , year=

    work page arXiv

  60. [65]

    Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , month =

    Zhu, Jie and Lee, Sungkil , title =. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , month =. 2025 , pages =

    work page 2025

  61. [66]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Mipi 2023 challenge on nighttime flare removal: Methods and results , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    work page 2023

  62. [67]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Light source guided single-image flare removal from unpaired data , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    work page

  63. [68]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Ff-former: Swin fourier transformer for nighttime flare removal , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    work page

  64. [69]

    arXiv preprint arXiv:2512.15221 , year=

    SLCFormer: Spectral-Local Context Transformer with Physics-Grounded Flare Synthesis for Nighttime Flare Removal , author=. arXiv preprint arXiv:2512.15221 , year=

    work page arXiv

  65. [70]

    IEEE Transactions on Circuits and Systems for Video Technology , year=

    LPFSformer: Location Prior Guided Frequency and Spatial Interactive Learning for Nighttime Flare Removal , author=. IEEE Transactions on Circuits and Systems for Video Technology , year=

    work page

  66. [71]

    ICASSP 2024-2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages=

    Flare-free vision: Empowering uformer with depth insights , author=. ICASSP 2024-2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages=. 2024 , organization=

    work page 2024

  67. [72]

    Entropy , volume=

    Flare removal model based on sparse-uformer networks , author=. Entropy , volume=. 2024 , publisher=

    work page 2024

  68. [73]

    The Visual Computer , volume=

    Mfdnet: Multi-frequency deflare network for efficient nighttime flare removal , author=. The Visual Computer , volume=. 2024 , publisher=

    work page 2024

  69. [74]

    Scientific Reports , volume=

    SMFR-Net: simple multi-domain flare removal network , author=. Scientific Reports , volume=. 2025 , publisher=

    work page 2025

  70. [75]

    arXiv preprint arXiv:2407.14746 , year=

    Difflare: Removing image lens flare with latent diffusion model , author=. arXiv preprint arXiv:2407.14746 , year=

    work page arXiv

  71. [76]

    Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , month =

    Tsai, Shr-Ruei and Chang, Wei-Cheng and Lee, Jie-Ying and Su, Chih-Hai and Liu, Yu-Lun , title =. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , month =. 2025 , pages =

    work page 2025

  72. [77]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Disentangle nighttime lens flares: self-supervised generation-based lens flare removal , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    work page

  73. [78]

    Proceedings of the IEEE/CVF international conference on computer vision , pages=

    Improving lens flare removal with general-purpose pipeline and multiple light sources recovery , author=. Proceedings of the IEEE/CVF international conference on computer vision , pages=

    work page

  74. [79]

    Engineering Applications of Artificial Intelligence , volume=

    Flare detection and detail compensation for nighttime flare removal , author=. Engineering Applications of Artificial Intelligence , volume=. 2026 , publisher=

    work page 2026

  75. [80]

    arXiv preprint arXiv:2412.08200 , year=

    GN-FR: Generalizable Neural Radiance Fields for Flare Removal , author=. arXiv preprint arXiv:2412.08200 , year=

    work page arXiv

  76. [81]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Uformer: A general u-shaped transformer for image restoration , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    work page

  77. [82]

    Proceedings of the IEEE/CVF international conference on computer vision , pages=

    Swin transformer: Hierarchical vision transformer using shifted windows , author=. Proceedings of the IEEE/CVF international conference on computer vision , pages=

    work page

  78. [83]

    Advances in neural information processing systems , volume=

    Mean teachers are better role models: Weight-averaged consistency targets improve semi-supervised deep learning results , author=. Advances in neural information processing systems , volume=

    work page

  79. [84]

    Advances in neural information processing systems , volume=

    Fixmatch: Simplifying semi-supervised learning with consistency and confidence , author=. Advances in neural information processing systems , volume=

    work page

  80. [85]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Self-training with noisy student improves imagenet classification , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    work page

Showing first 80 references.