pith. sign in

arxiv: 2604.15903 · v1 · submitted 2026-04-17 · 💻 cs.CV

AeroDeshadow: Physics-Guided Shadow Synthesis and Penumbra-Aware Deshadowing for Aerospace Imagery

Wei Lu , Zi-Yang Bo , Fei-Fei Sang , Yi Liu , Xue Yang , Si-Bao Chen This is my paper

Pith reviewed 2026-05-10 08:29 UTC · model grok-4.3

classification 💻 cs.CV
keywords shadow removalaerospace imagerypenumbraphysics-guided synthesisimage restorationsynthetic dataset generationdeep learning
0
0 comments X

The pith

Physics-guided synthesis of soft-shadow pairs lets a deshadowing network generalize from synthetic aerospace data to real images without paired real annotations.

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

The paper shows that an explicit model of illumination decay and spatial attenuation can generate large-scale paired shadow data featuring realistic soft boundaries. This synthetic dataset then trains a cascaded restoration network that separates umbra and penumbra processing to reduce boundary artifacts and spectral distortion. The resulting model, trained only on the synthetic pairs, produces state-of-the-art shadow removal on both synthetic test sets and unpaired real aerospace imagery.

Core claim

AeroDeshadow integrates a Physics-aware Degradation Shadow Synthesis Network that constructs paired data by modeling physical illumination decay and attenuation, with a Penumbra-aware Cascaded DeShadowing Network that progressively restores umbra and penumbra regions, allowing the full system to generalize from the synthetic AeroDS-Syn dataset to real aerospace scenes.

What carries the argument

Physics-aware Degradation Shadow Synthesis Network that explicitly models illumination decay and spatial attenuation to produce paired data with soft penumbra transitions for training the restoration stage.

Load-bearing premise

The chosen model of illumination decay and spatial attenuation accurately reproduces the broad penumbra zones and spectral changes found in real aerospace imagery.

What would settle it

Direct measurement of penumbra width and intensity fall-off profiles on real high-resolution aerospace images that deviate substantially from the profiles generated by the synthesis network would falsify the claim that the synthetic data supports reliable generalization.

Figures

Figures reproduced from arXiv: 2604.15903 by Fei-Fei Sang, Si-Bao Chen, Wei Lu, Xue Yang, Yi Liu, Zi-Yang Bo.

Figure 1
Figure 1. Figure 1: Visual comparison of shadow removal on synthetic (top) and real-world (bottom) aerospace imagery from the proposed AeroDS dataset. The boxes [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overall architecture of the proposed AeroDeshadow framework. The [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Geographical distribution and representative samples of AeroDS. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Representative samples from AeroDS-Syn and AeroDS-Real. [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Detailed flowchart of the AeroDS-Syn dataset construction. The process integrates a Shadow Illumination Decay Estimator to extract physical [PITH_FULL_IMAGE:figures/full_fig_p005_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Detailed architecture of the SDCA module. To model the physical [PITH_FULL_IMAGE:figures/full_fig_p005_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Architecture of the PCDS-Net. The model explicitly decouples the input into umbra and penumbra streams. An Attention Feature Fusion (AFF) module [PITH_FULL_IMAGE:figures/full_fig_p007_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Qualitative comparison of shadow synthesis methods, accompanied by pixel intensity profile analysis across shadow boundaries. The profile curves [PITH_FULL_IMAGE:figures/full_fig_p009_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Qualitative comparison of shadow removal performance on the AeroDS-Syn (top two rows) and SRGTA (bottom two rows) datasets. Boxes indicate [PITH_FULL_IMAGE:figures/full_fig_p010_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Qualitative comparison of shadow removal performance on real-world ASI from the AeroDS-Real (top two rows) and AISD (bottom two rows) [PITH_FULL_IMAGE:figures/full_fig_p011_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Visual ablation analysis of the SDCA module within PDSS-Net. The [PITH_FULL_IMAGE:figures/full_fig_p011_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Qualitative visual comparison of the ablation study for the proposed [PITH_FULL_IMAGE:figures/full_fig_p012_12.png] view at source ↗
read the original abstract

Shadows are prevalent in high-resolution aerospace imagery (ASI). They often cause spectral distortion and information loss, which degrade downstream interpretation tasks. While deep learning methods have advanced natural-image shadow removal, their direct application to ASI faces two primary challenges. First, strictly paired training data are severely lacking. Second, homogeneous shadow assumptions fail to handle the broad penumbra transition zones inherent in aerospace scenes. To address these issues, we propose AeroDeshadow, a unified two-stage framework integrating physics-guided shadow synthesis and penumbra-aware restoration. In the first stage, a Physics-aware Degradation Shadow Synthesis Network (PDSS-Net) explicitly models illumination decay and spatial attenuation. This process constructs AeroDS-Syn, a large-scale paired dataset featuring soft boundary transitions. Constrained by this physical formulation, a Penumbra-aware Cascaded DeShadowing Network (PCDS-Net) then decouples the input into umbra and penumbra components. By restoring these regions progressively, PCDS-Net alleviates boundary artifacts and over-correction. Trained solely on the synthetic AeroDS-Syn, the network generalizes to real-world ASI without requiring paired real annotations. Experimental results indicate that AeroDeshadow achieves state-of-the-art quantitative accuracy and visual fidelity across synthetic and real-world datasets. The datasets and code will be made publicly available at: https://github.com/AeroVILab-AHU/AeroDeshadow.

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

3 major / 2 minor

Summary. The paper proposes AeroDeshadow, a two-stage framework for shadow removal in high-resolution aerospace imagery (ASI). PDSS-Net explicitly models illumination decay and spatial attenuation to synthesize the paired AeroDS-Syn dataset featuring soft penumbra boundaries. PCDS-Net then decouples umbra and penumbra regions for progressive restoration. Trained exclusively on synthetic data, the method claims to generalize to real unpaired ASI and achieve SOTA quantitative accuracy plus visual fidelity on both synthetic and real datasets, with public release of data and code.

Significance. If the synthetic physics model faithfully reproduces real ASI penumbra transitions and spectral effects, the work would meaningfully address paired-data scarcity in aerospace shadow removal and enable more robust downstream tasks. The public dataset and code release strengthens reproducibility. At present, however, the unvalidated match between synthetic and real statistics substantially reduces the assessed significance of the generalization result.

major comments (3)
  1. [§3.1] §3.1 (PDSS-Net description): The explicit illumination decay and spatial attenuation formulation is presented as constructing soft-boundary AeroDS-Syn, yet no quantitative fidelity metrics (boundary-gradient histograms, shadowed-region reflectance distributions, or atmospheric scattering parameter matches) are reported against real ASI statistics. This match is load-bearing for the abstract's generalization claim that synthetic-only training yields reliable real-world performance.
  2. [§4] §4 (Experiments) and abstract: The SOTA quantitative accuracy claim on real-world datasets is stated without accompanying tables of metrics (PSNR, SSIM, or error distributions), ablation studies on the physics components, or error analysis; only the synthetic dataset is described as having quantitative support. This leaves the central generalization assertion without verifiable evidence.
  3. [§3.2] §3.2 (PCDS-Net): The progressive umbra/penumbra decoupling is motivated by homogeneous-shadow limitations, but no ablation isolating the penumbra-aware cascade versus a standard shadow-removal backbone is provided to demonstrate that the added components are responsible for the reported boundary-artifact reduction.
minor comments (2)
  1. Notation for the illumination decay and attenuation parameters in PDSS-Net could be introduced with a single consolidated table of symbols and physical units.
  2. Figure 1 (framework overview) would benefit from explicit call-outs labeling the umbra versus penumbra regions in both synthetic and real examples.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the thorough and constructive review of our manuscript. We address each of the major comments below and have revised the paper accordingly to strengthen the presentation of our results.

read point-by-point responses

  1. Referee: [§3.1] The explicit illumination decay and spatial attenuation formulation is presented as constructing soft-boundary AeroDS-Syn, yet no quantitative fidelity metrics (boundary-gradient histograms, shadowed-region reflectance distributions, or atmospheric scattering parameter matches) are reported against real ASI statistics. This match is load-bearing for the abstract's generalization claim that synthetic-only training yields reliable real-world performance.

    Authors: We agree that providing quantitative fidelity metrics would better substantiate the physical realism of the synthesized data and support the generalization claims. In the revised manuscript, we have added quantitative comparisons in §3.1, including boundary-gradient histograms and reflectance distribution statistics between the synthetic AeroDS-Syn and real ASI samples. For atmospheric scattering parameters, we have included a sensitivity analysis and qualitative matches to established models, as direct parameter extraction from real imagery requires additional instrumentation not available in our current dataset. revision: partial

  2. Referee: [§4] The SOTA quantitative accuracy claim on real-world datasets is stated without accompanying tables of metrics (PSNR, SSIM, or error distributions), ablation studies on the physics components, or error analysis; only the synthetic dataset is described as having quantitative support. This leaves the central generalization assertion without verifiable evidence.

    Authors: We acknowledge the need for clarity here. Quantitative metrics such as PSNR and SSIM are only feasible on the synthetic dataset where paired ground truth exists. For real-world ASI, we have revised the abstract and §4 to remove the 'quantitative accuracy' claim for real data and instead highlight visual fidelity, perceptual quality, and successful generalization as evidenced by qualitative results and downstream task improvements. We have also added ablation studies on the physics components and error analysis on the synthetic test set to provide more verifiable evidence. revision: yes

  3. Referee: [§3.2] The progressive umbra/penumbra decoupling is motivated by homogeneous-shadow limitations, but no ablation isolating the penumbra-aware cascade versus a standard shadow-removal backbone is provided to demonstrate that the added components are responsible for the reported boundary-artifact reduction.

    Authors: We appreciate this observation. To demonstrate the effectiveness of the penumbra-aware cascade, we have included a new ablation study in §4. This compares the full PCDS-Net against a baseline using a standard shadow-removal network without the umbra-penumbra decoupling and progressive restoration. The results confirm that the proposed components significantly reduce boundary artifacts and improve restoration quality. revision: yes

Circularity Check

0 steps flagged

No circularity: new physics-guided synthesis and cascaded restoration networks are independent of target performance

full rationale

The paper's chain introduces PDSS-Net to generate AeroDS-Syn via explicit illumination decay and spatial attenuation modeling, then trains PCDS-Net on that synthetic data for progressive umbra/penumbra restoration. No equations, fitted parameters, or self-citations are shown that reduce the claimed generalization or SOTA metrics to a re-expression of the inputs by construction. The physics model and network architectures are presented as novel contributions rather than tautological renamings or fitted-input predictions. The central claim (synthetic-only training yields real-world performance) rests on an external assumption of model fidelity, which is a validation gap rather than a definitional loop.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on an unstated but load-bearing assumption that a parametric model of illumination decay and spatial attenuation can generate synthetic shadows whose statistics match real aerospace penumbras sufficiently well for generalization. No free parameters are named in the abstract, but the physics model necessarily introduces at least scale and attenuation coefficients. No new physical entities are postulated.

axioms (1)
  • domain assumption A parametric model of illumination decay and spatial attenuation can produce synthetic shadow pairs whose penumbra statistics are close enough to real aerospace imagery to support generalization.
    Invoked in the description of PDSS-Net and the claim that training solely on AeroDS-Syn enables real-world performance.

pith-pipeline@v0.9.0 · 5567 in / 1471 out tokens · 32769 ms · 2026-05-10T08:29:16.078474+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

55 extracted references · 55 canonical work pages

  1. [1]

    Deep learning in remote sensing: A comprehensive review and list of resources,

    X. X. Zhu, D. Tuia, L. Mou, G.-S. Xia, L. Zhang, F. Xu, and F. Fraundorfer, “Deep learning in remote sensing: A comprehensive review and list of resources,”IEEE Geosci. Remote Sens. Mag., vol. 5, no. 4, pp. 8–36, 2017

    work page 2017

  2. [2]

    Remote sensing image scene classifica- tion: Benchmark and state of the art,

    G. Cheng, J. Han, and X. Lu, “Remote sensing image scene classifica- tion: Benchmark and state of the art,”Proc. IEEE, vol. 105, no. 10, pp. 1865–1883, 2017

    work page 2017

  3. [3]

    A robust feature downsampling module for remote-sensing visual tasks,

    W. Lu, S.-B. Chen, J. Tang, C. H. Ding, and B. Luo, “A robust feature downsampling module for remote-sensing visual tasks,”IEEE Trans. Geosci. Remote Sens., vol. 61, pp. 1–12, 2023

    work page 2023

  4. [4]

    Attention-aware sobel graph convolutional network for remote sensing image change detection,

    L. Wang, Z. You, W. Lu, S. Chen, J. Tang, and B. Luo, “Attention-aware sobel graph convolutional network for remote sensing image change detection,”IEEE Trans. Geosci. Remote Sens., vol. 62, pp. 1–12, 2024

    work page 2024

  5. [5]

    UniRoute: Unified routing mixture-of-experts for modality-adaptive remote sensing change detection,

    Q. Shu, S. Chen, W. Lu, Z. You, and C. Liu, “UniRoute: Unified routing mixture-of-experts for modality-adaptive remote sensing change detection,”arXiv preprint arXiv:2601.14797, 2026

    work page arXiv 2026

  6. [6]

    Shadow detection in colour high-resolution satellite images,

    V . Ar´evalo, J. Gonz´alez, and G. Ambrosio, “Shadow detection in colour high-resolution satellite images,”Int. J. Remote Sens., vol. 29, no. 7, pp. 1945–1963, 2008

    work page 1945

  7. [7]

    Real- world remote sensing image dehazing: Benchmark and baseline,

    Z.-H. Zhu, W. Lu, S.-B. Chen, C. H. Ding, J. Tang, and B. Luo, “Real- world remote sensing image dehazing: Benchmark and baseline,”IEEE Trans. Geosci. Remote Sens., vol. 63, pp. 1–14, 2025

    work page 2025

  8. [8]

    LEGNet: A lightweight edge-Gaussian network for low-quality remote sensing image object detection,

    W. Lu, S.-B. Chen, H.-D. Li, Q.-L. Shu, C. H. Ding, J. Tang, and B. Luo, “LEGNet: A lightweight edge-Gaussian network for low-quality remote sensing image object detection,” inProc. IEEE Conf. Int. Conf. Comput. Vis. Worksh., 2025, pp. 2844–2853

    work page 2025

  9. [9]

    Perceive-IR: Learning to perceive degradation better for all-in-one image restoration,

    X. Zhang, J. Ma, G. Wang, Q. Zhang, H. Zhang, and L. Zhang, “Perceive-IR: Learning to perceive degradation better for all-in-one image restoration,”IEEE Trans. Image Process., vol. 35, pp. 2018–2033, 2026

    work page 2018

  10. [10]

    UniUIR: Considering underwater image restoration as an all-in-one learner,

    X. Zhang, H. Zhang, G. Wang, Q. Zhang, L. Zhang, and B. Du, “UniUIR: Considering underwater image restoration as an all-in-one learner,”IEEE Trans. Image Process., vol. 34, pp. 6963–6977, 2025

    work page 2025

  11. [11]

    UnravelNet: A backbone for enhanced multi-scale and low- quality feature extraction in remote sensing object detection,

    W. Lu, H.-D. Li, C. Wang, S.-B. Chen, C. H. Ding, J. Tang, and B. Luo, “UnravelNet: A backbone for enhanced multi-scale and low- quality feature extraction in remote sensing object detection,”ISPRS J. Photogramm. Remote Sens., vol. 231, pp. 431–442, 2026

    work page 2026

  12. [12]

    DOTA: A large-scale dataset for object detection in aerial images,

    G.-S. Xia, X. Bai, J. Ding, Z. Zhu, S. J. Belongie, J. Luo, M. Datcu, M. Pelillo, and L. Zhang, “DOTA: A large-scale dataset for object detection in aerial images,” inProc. IEEE Conf. Comput. Vis. Pattern Recog., 2018, pp. 3974–3983

    work page 2018

  13. [13]

    DecoupleNet: A lightweight backbone network with efficient feature decoupling for remote sensing visual tasks,

    W. Lu, S.-B. Chen, Q.-L. Shu, J. Tang, and B. Luo, “DecoupleNet: A lightweight backbone network with efficient feature decoupling for remote sensing visual tasks,”IEEE Trans. Geosci. Remote Sens., vol. 62, pp. 1–13, 2024

    work page 2024

  14. [14]

    LWGANet: Addressing spatial and channel redundancy in remote sensing visual tasks with light-weight grouped attention,

    W. Lu, X. Yang, and S.-B. Chen, “LWGANet: Addressing spatial and channel redundancy in remote sensing visual tasks with light-weight grouped attention,” inProc. AAAI Conf. Artif. Intell., vol. 40, no. 9, 2026, pp. 7574–7582

    work page 2026

  15. [15]

    On the removal of shadows from images,

    G. D. Finlayson, S. D. Hordley, C. Lu, and M. S. Drew, “On the removal of shadows from images,”IEEE Trans. Pattern Anal. Mach. Intell., vol. 28, no. 1, pp. 59–68, 2006

    work page 2006

  16. [16]

    Physics-based shadow image decomposition for shadow removal,

    H. Le and D. Samaras, “Physics-based shadow image decomposition for shadow removal,”IEEE Trans. Pattern Anal. Mach. Intell., vol. 44, no. 12, pp. 9088–9101, 2022

    work page 2022

  17. [17]

    Paired regions for shadow detection and removal,

    R. Guo, Q. Dai, and D. Hoiem, “Paired regions for shadow detection and removal,”IEEE Trans. Pattern Anal. Mach. Intell., vol. 35, no. 12, pp. 2956–2967, 2013

    work page 2013

  18. [18]

    Shadow detection and removal from a single image using lab color space,

    S. Murali and V . Govindan, “Shadow detection and removal from a single image using lab color space,”Cybernetics and information technologies, vol. 13, no. 1, pp. 95–103, 2013

    work page 2013

  19. [19]

    DeshadowNet: A multi- context embedding deep network for shadow removal,

    L. Qu, J. Tian, S. He, Y . Tang, and R. W. Lau, “DeshadowNet: A multi- context embedding deep network for shadow removal,” inProc. IEEE Conf. Comput. Vis. Pattern Recog., 2017, pp. 4067–4075

    work page 2017

  20. [20]

    Stacked conditional generative adversarial networks for jointly learning shadow detection and shadow removal,

    J. Wang, X. Li, and J. Yang, “Stacked conditional generative adversarial networks for jointly learning shadow detection and shadow removal,” in Proc. IEEE Conf. Comput. Vis. Pattern Recog., 2018, pp. 1788–1797

    work page 2018

  21. [21]

    CANet: A context-aware network for shadow removal,

    Z. Chen, C. Long, L. Zhang, and C. Xiao, “CANet: A context-aware network for shadow removal,” inProc. IEEE Conf. Int. Conf. Comput. Vis., 2021, pp. 4743–4752

    work page 2021

  22. [22]

    ShadowDiffusion: When degradation prior meets diffusion model for shadow removal,

    L. Guo, C. Wang, W. Yang, S. Huang, Y . Wang, H. Pfister, and B. Wen, “ShadowDiffusion: When degradation prior meets diffusion model for shadow removal,” inProc. IEEE Conf. Comput. Vis. Pattern Recog., 2023, pp. 14 049–14 058. IEEE TRANSACTIONS ON IMAGE PROCESSING, 2026 13

    work page 2023

  23. [23]

    An evolutionary shadow correction network and a benchmark uav dataset for remote sensing images,

    S. Luo, H. Li, Y . Li, C. Shao, H. Shen, and L. Zhang, “An evolutionary shadow correction network and a benchmark uav dataset for remote sensing images,”IEEE Trans. Geosci. Remote Sens., vol. 61, pp. 1–14, 2023

    work page 2023

  24. [24]

    RMMamba: Randomized mamba for remote sensing shadow removal,

    J. Chu, K. Chi, and Q. Wang, “RMMamba: Randomized mamba for remote sensing shadow removal,”IEEE Trans. Geosci. Remote Sens., vol. 63, pp. 1–10, 2025

    work page 2025

  25. [25]

    Boundary-aware divide and conquer: A diffusion-based solution for unsupervised shadow removal,

    L. Guo, C. Wang, W. Yang, Y . Wang, and B. Wen, “Boundary-aware divide and conquer: A diffusion-based solution for unsupervised shadow removal,” inProc. IEEE Conf. Int. Conf. Comput. Vis., 2023, pp. 13 045– 13 054

    work page 2023

  26. [26]

    Recreating brightness from remote sensing shadow appearance,

    Q. Wang, K. Chi, W. Jing, and Y . Yuan, “Recreating brightness from remote sensing shadow appearance,”IEEE Trans. Geosci. Remote Sens., vol. 62, pp. 1–11, 2024

    work page 2024

  27. [27]

    Shadow removal by a lightness-guided network with training on unpaired data,

    Z. Liu, H. Yin, Y . Mi, M. Pu, and S. Wang, “Shadow removal by a lightness-guided network with training on unpaired data,”IEEE Trans. Image Process., vol. 30, pp. 1853–1865, 2021

    work page 2021

  28. [28]

    Unsupervised ultra-high-resolution uav low-light image enhancement: A benchmark, metric and framework,

    W. Lu, L. Zhu, and S.-B. Chen, “Unsupervised ultra-high-resolution uav low-light image enhancement: A benchmark, metric and framework,” arXiv preprint arXiv:2509.01373, 2025

    work page arXiv 2025

  29. [29]

    DC-ShadowNet: Single-image hard and soft shadow removal using unsupervised domain-classifier guided network,

    Y . Jin, A. Sharma, and R. T. Tan, “DC-ShadowNet: Single-image hard and soft shadow removal using unsupervised domain-classifier guided network,” inProc. IEEE Conf. Int. Conf. Comput. Vis., 2021, pp. 5027– 5036

    work page 2021

  30. [30]

    A method to generate soft shadows using a layered depth image and warping,

    Y .-H. Im, C.-Y . Han, and L.-S. Kim, “A method to generate soft shadows using a layered depth image and warping,”IEEE Trans. Vis. Comput. Graph., vol. 11, no. 3, pp. 265–272, 2005

    work page 2005

  31. [31]

    Unpaired image-to-image translation using cycle-consistent adversarial networks,

    J.-Y . Zhu, T. Park, P. Isola, and A. A. Efros, “Unpaired image-to-image translation using cycle-consistent adversarial networks,” inProc. IEEE Conf. Int. Conf. Comput. Vis., 2017, pp. 2242–2251

    work page 2017

  32. [32]

    Mask-ShadowGAN: Learning to remove shadows from unpaired data,

    X. Hu, Y . Jiang, C.-W. Fu, and P.-A. Heng, “Mask-ShadowGAN: Learning to remove shadows from unpaired data,” inProc. IEEE Conf. Int. Conf. Comput. Vis., 2019, pp. 2472–2481

    work page 2019

  33. [33]

    From shadow generation to shadow removal,

    Z. Liu, H. Yin, X. Wu, Z. Wu, Y . Mi, and S. Wang, “From shadow generation to shadow removal,” inProc. IEEE Conf. Comput. Vis. Pattern Recog., 2021, pp. 4925–4934

    work page 2021

  34. [34]

    Shadow generation using diffusion model with geometry prior,

    H. Zhao, Q. Liu, X. Tao, L. Niu, and G. Zhai, “Shadow generation using diffusion model with geometry prior,” inProc. IEEE Conf. Comput. Vis. Pattern Recog., 2025, pp. 7603–7612

    work page 2025

  35. [35]

    Auto-exposure fusion for single-image shadow removal,

    L. Fu, C. Zhou, Q. Guo, F. Juefei-Xu, H. Yu, W. Feng, Y . Liu, and S. Wang, “Auto-exposure fusion for single-image shadow removal,” in Proc. IEEE Conf. Comput. Vis. Pattern Recog., 2021, pp. 10 566–10 575

    work page 2021

  36. [36]

    Auto- matic shadow generation via exposure fusion,

    Q. Meng, S. Zhang, Z. Li, C. Wang, W. Zhang, and Q. Huang, “Auto- matic shadow generation via exposure fusion,”IEEE Trans. Multimedia, vol. 25, pp. 9044–9056, 2023

    work page 2023

  37. [37]

    From synthesis to removal: A deep learning-based framework for shadow removal in high-resolution remote sensing images,

    C. Shao, H. Li, L. Xu, M. Gao, and H. Shen, “From synthesis to removal: A deep learning-based framework for shadow removal in high-resolution remote sensing images,” inInt’l Geosci. Remote Sens. Symp., 2024, pp. 7488–7491

    work page 2024

  38. [39]

    Shadow remover: Image shadow removal based on illumination recovering optimization,

    L. Zhang, Q. Zhang, and C. Xiao, “Shadow remover: Image shadow removal based on illumination recovering optimization,”IEEE Trans. Image Process., vol. 24, no. 11, pp. 4623–4636, 2015

    work page 2015

  39. [40]

    WSRD: A novel bench- mark for high resolution image shadow removal,

    F.-A. Vasluianu, T. Seizinger, and R. Timofte, “WSRD: A novel bench- mark for high resolution image shadow removal,” inProc. IEEE Conf. Comput. Vis. Pattern Recog. Worksh., 2023, pp. 1826–1835

    work page 2023

  40. [41]

    ShadowFormer: Global context helps shadow removal,

    L. Guo, S. Huang, D. Liu, H. Cheng, and B. Wen, “ShadowFormer: Global context helps shadow removal,”Proc. AAAI Conf. Artif. Intell., vol. 37, no. 1, pp. 710–718, Jun. 2023

    work page 2023

  41. [42]

    HomoFormer: Homogenized transformer for image shadow removal,

    J. Xiao, X. Fu, Y . Zhu, D. Li, J. Huang, K. Zhu, and Z.-J. Zha, “HomoFormer: Homogenized transformer for image shadow removal,” inProc. IEEE Conf. Comput. Vis. Pattern Recog., 2024, pp. 25 617– 25 626

    work page 2024

  42. [43]

    HirFormer: Dynamic high resolution transformer for large-scale image shadow removal,

    X. Lu, Y . Zhu, X. Wang, D. Li, J. Xiao, Y . Zhang, X. Fu, and Z.-J. Zha, “HirFormer: Dynamic high resolution transformer for large-scale image shadow removal,” inProc. IEEE Conf. Comput. Vis. Pattern Recog. Worksh., 2024, pp. 6513–6523

    work page 2024

  43. [44]

    Learning to remove shadows from a single image,

    H. Jiang, Q. Zhang, Y . Nie, L. Zhu, and W.-S. Zheng, “Learning to remove shadows from a single image,”Int. J. Comput. Vis., vol. 131, no. 9, pp. 2471–2488, 2023

    work page 2023

  44. [45]

    Latent feature-guided diffusion models for shadow removal,

    K. Mei, L. Figueroa, Z. Lin, Z. Ding, S. Cohen, and V . M. Patel, “Latent feature-guided diffusion models for shadow removal,” inIEEE Winter Conf. Appl. Comput. Vis., 2024, pp. 4301–4310

    work page 2024

  45. [46]

    Diff-Shadow: Global-guided diffusion model for shadow removal,

    J. Luo, R. Li, C. Jiang, X. Zhang, M. Han, T. Jiang, H. Fan, and S. Liu, “Diff-Shadow: Global-guided diffusion model for shadow removal,” Proc. AAAI Conf. Artif. Intell., vol. 39, no. 6, pp. 5856–5864, 2025

    work page 2025

  46. [47]

    Semantic-Guided coarse-to-fine diffusion model for self-supervised image shadow removal,

    Z. Zeng, C. Zhao, W. Cai, and Y . Guo, “Semantic-Guided coarse-to-fine diffusion model for self-supervised image shadow removal,” inComput. Vis. Media, 2025, pp. 271–293

    work page 2025

  47. [48]

    RSMamba: Biologically plausible retinex-based mamba for remote sensing shadow removal,

    K. Chi, S. Guo, J. Chu, Q. Li, and Q. Wang, “RSMamba: Biologically plausible retinex-based mamba for remote sensing shadow removal,” IEEE Trans. Geosci. Remote Sens., vol. 63, pp. 1–10, 2025

    work page 2025

  48. [49]

    Real-ESRGAN: Training real- world blind super-resolution with pure synthetic data,

    X. Wang, L. Xie, C. Dong, and Y . Shan, “Real-ESRGAN: Training real- world blind super-resolution with pure synthetic data,” inProc. IEEE Conf. Int. Conf. Comput. Vis. Worksh., 2021, pp. 1905–1914

    work page 2021

  49. [50]

    U-Net: Convolutional net- works for biomedical image segmentation,

    O. Ronneberger, P. Fischer, and T. Brox, “U-Net: Convolutional net- works for biomedical image segmentation,” inInternational Conference on Medical image computing and computer-assisted intervention, 2015, pp. 234–241

    work page 2015

  50. [51]

    Deeply supervised convolutional neural network for shadow detection based on a novel aerial shadow imagery dataset,

    S. Luo, H. Li, and H. Shen, “Deeply supervised convolutional neural network for shadow detection based on a novel aerial shadow imagery dataset,”ISPRS J. Photogramm. Remote Sens., vol. 167, pp. 443–457, 2020

    work page 2020

  51. [52]

    Learning from Synthetic Shadows for Shadow Detection and Removal,

    N. Inoue and T. Yamasaki, “Learning from Synthetic Shadows for Shadow Detection and Removal,”IEEE Trans. Circuit Syst. Video Technol., vol. 31, no. 11, pp. 4187–4197, 2021

    work page 2021

  52. [53]

    Generative shadow synthesis and removal for remote sensing images through embedding illumination models,

    C. Shao, H. Li, and H. Shen, “Generative shadow synthesis and removal for remote sensing images through embedding illumination models,” IEEE Trans. Geosci. Remote Sens., vol. 63, pp. 1–15, 2025

    work page 2025

  53. [54]

    Near real-time shadow detection and removal in aerial motion imagery application,

    G. F. Silva, G. B. Carneiro, R. Doth, L. A. Amaral, and D. F. de Azevedo, “Near real-time shadow detection and removal in aerial motion imagery application,”ISPRS J. Photogramm. Remote Sens., vol. 140, pp. 104– 121, 2018

    work page 2018

  54. [55]

    Shadow detection and removal for remote sensing images via multi-feature adaptive optimization and geometry-aware illumination compensation,

    Z. Zhang, R. Cao, H. Sheng, M. Guo, Z. Shao, and L. Wu, “Shadow detection and removal for remote sensing images via multi-feature adaptive optimization and geometry-aware illumination compensation,” Expert Syst. Appl., vol. 282, p. 127769, 2025

    work page 2025

  55. [56]

    Image quality assessment: from error visibility to structural similarity,

    Z. Wang, A. C. Bovik, H. R. Sheikh, and E. P. Simoncelli, “Image quality assessment: from error visibility to structural similarity,”IEEE Trans. Image Process., vol. 13, no. 4, pp. 600–612, 2004

    work page 2004