pith. sign in

arxiv: 2604.04136 · v1 · submitted 2026-04-05 · 💻 cs.CV

Rethinking Exposure Correction for Spatially Non-uniform Degradation

Ao Li , Jiawei Sun , Le Dong , Zhenyu Wang , Weisheng Dong This is my paper

Pith reviewed 2026-05-13 16:47 UTC · model grok-4.3

classification 💻 cs.CV
keywords exposure correctionnon-uniform degradationspatial signal encoderlook-up tablesuncertainty lossimage enhancementHSL compensationreal-world photos
0
0 comments X

The pith

Spatially adaptive modulation weights and an uncertainty loss correct non-uniform exposure degradations better than global methods.

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

The paper contends that real-world exposure errors vary across regions within one image, yet prior correction techniques aggregate signals globally and apply uniform loss scales that ignore those differences. It introduces a Spatial Signal Encoder to generate local modulation weights that drive multiple look-up tables for transformation, plus an HSL compensation module and a loss that shifts focus toward locally uncertain pixels. Experiments on standard benchmarks show improved qualitative detail and quantitative scores over existing approaches. A reader would care because many photos contain mixed bright and dark zones where uniform fixes leave visible artifacts.

Core claim

The central claim is that a new exposure correction architecture explicitly built for spatial non-uniformity outperforms prior work: a Spatial Signal Encoder predicts adaptive modulation weights to control multiple look-up tables, an HSL-based module restores color fidelity, and an uncertainty-inspired loss dynamically weights optimization according to local restoration difficulty, producing superior results on real images containing heterogeneous exposure errors.

What carries the argument

Spatial Signal Encoder that outputs spatially varying modulation weights to steer multiple look-up tables, paired with an uncertainty-inspired non-uniform loss that reallocates gradients to high-uncertainty regions.

If this is right

  • Regions with simultaneous over- and under-exposure receive targeted corrections instead of averaged compromises.
  • Color shifts are reduced through the separate HSL compensation step in heterogeneous lighting.
  • Optimization automatically emphasizes harder local patches without manual region masks.
  • The approach produces measurable lifts in PSNR, SSIM, and perceptual metrics on real-world photo datasets.
  • Qualitative outputs exhibit fewer halo or banding artifacts at exposure boundaries.

Where Pith is reading between the lines

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

  • The same local-modulation idea could be tested on other spatially varying degradations such as motion blur or sensor noise.
  • End-to-end integration with segmentation or depth networks might further refine the modulation weights.
  • Extending the loss to video frames could expose whether temporal consistency improves or requires extra constraints.
  • The uncertainty weighting mechanism suggests a general template for any restoration task where error distribution is non-stationary.

Load-bearing premise

Existing methods are limited mainly because they rely on globally aggregated signals and shared scale losses that miss spatially varying correction needs.

What would settle it

Performance on a controlled test set of uniformly exposed images would match or fall below current global methods, or ablating the spatial encoder on non-uniform data would eliminate the reported gains.

Figures

Figures reproduced from arXiv: 2604.04136 by Ao Li, Jiawei Sun, Le Dong, Weisheng Dong, Zhenyu Wang.

Figure 1
Figure 1. Figure 1: Motivation for Our Methodology. and the required restoration is usually close to global brightening. In contrast, for high-dynamic-range scenes, even modern cameras with automatic exposure control often fail to capture the full scene dynamic range due to the limited sensing range of the camera sensor. As a result, only part of the scene, typically the metered region, may be properly exposed, while other re… view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the proposed framework. Given an incorrectly exposed input image, our method first performs non [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Illustration of the proposed HSL-based compensa [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Visual comparisons with state-of-the-art methods on the MSEC dataset [1]. [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Visual comparisons with state-of-the-art methods on the SICE dataset [5]. [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Visual comparisons with state-of-the-art methods on the LCDP dataset [37]. [PITH_FULL_IMAGE:figures/full_fig_p007_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Visual comparison of different modulation strate [PITH_FULL_IMAGE:figures/full_fig_p008_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Quantitative and structural comparison of differ [PITH_FULL_IMAGE:figures/full_fig_p008_8.png] view at source ↗
read the original abstract

Real-world exposure correction is fundamentally challenged by spatially non-uniform degradations, where diverse exposure errors frequently coexist within a single image. However, existing exposure correction methods are still largely developed under a predominantly uniform assumption. Architecturally, they typically rely on globally aggregated modulation signals that capture only the overall exposure trend. From the optimization perspective, conventional reconstruction losses are usually derived under a shared global scale, thus overlooking the spatially varying correction demands across regions. To address these limitations, we propose a new exposure correction paradigm explicitly designed for spatial non-uniformity. Specifically, we introduce a Spatial Signal Encoder to predict spatially adaptive modulation weights, which are used to guide multiple look-up tables for image transformation, together with an HSL-based compensation module for improved color fidelity. Beyond the architectural design, we propose an uncertainty-inspired non-uniform loss that dynamically allocates the optimization focus based on local restoration uncertainties, better matching the heterogeneous nature of real-world exposure errors. Extensive experiments demonstrate that our method achieves superior qualitative and quantitative performance compared with state-of-the-art methods. Code is available at https://github.com/FALALAS/rethinkingEC.

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

0 major / 2 minor

Summary. The paper proposes a new exposure correction paradigm for handling spatially non-uniform degradations in real-world images. It introduces a Spatial Signal Encoder to predict adaptive modulation weights that guide multiple look-up tables for image transformation, an HSL-based compensation module for color fidelity, and an uncertainty-inspired non-uniform loss that dynamically weights regions based on local restoration uncertainties. The authors claim this addresses limitations of global modulation and shared-scale losses in prior methods, with extensive experiments showing superior qualitative and quantitative performance over state-of-the-art approaches. Code is provided for reproducibility.

Significance. If the performance claims hold under full verification, the work meaningfully advances exposure correction by explicitly targeting spatial variation, a common real-world challenge overlooked by global-assumption methods. The architectural choices (adaptive modulation via encoder and uncertainty-weighted loss) are well-aligned with the problem, and code availability supports independent validation and extension. This could influence downstream applications in computational photography and image restoration.

minor comments (2)
  1. The abstract and introduction would benefit from a brief quantitative comparison (e.g., PSNR/SSIM deltas) against the strongest baseline to ground the 'superior performance' claim before the detailed experiments section.
  2. Notation for the modulation weights and uncertainty map should be defined consistently in the method section (e.g., clarify whether the Spatial Signal Encoder outputs are normalized per-channel or globally).

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the positive summary, significance assessment, and recommendation of minor revision. We are encouraged that the spatially adaptive design and uncertainty-inspired loss are recognized as well-aligned with real-world non-uniform exposure correction. No major comments were provided in the report, so we have no specific points to address point-by-point. We will incorporate any minor suggestions during revision and ensure the code remains available for verification.

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper introduces a Spatial Signal Encoder for adaptive modulation weights, an HSL compensation module, and an uncertainty-inspired non-uniform loss to target spatially varying exposure errors. These are presented as new architectural and loss-function choices rather than derivations that reduce outputs to inputs by construction. No equations equate predictions to fitted parameters or prior self-citations in a load-bearing way. Performance superiority is asserted via experiments on standard benchmarks, which remain externally falsifiable. Code release provides an independent verification path. The derivation chain is self-contained against external benchmarks with no self-definitional, fitted-input, or uniqueness-imported circular steps.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 1 invented entities

Abstract-only review prevents enumeration of specific fitted values or background axioms; the work introduces new architectural modules without external benchmarks or formal proofs.

invented entities (1)
  • Spatial Signal Encoder no independent evidence
    purpose: Predict spatially adaptive modulation weights to guide lookup tables
    New component introduced to address non-uniformity; no independent evidence provided beyond the method itself.

pith-pipeline@v0.9.0 · 5497 in / 1051 out tokens · 43844 ms · 2026-05-13T16:47:27.787994+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

51 extracted references · 51 canonical work pages · 1 internal anchor

  1. [1]

    Mahmoud Afifi, Konstantinos G Derpanis, Bjorn Ommer, and Michael S Brown

    work page

  2. [2]

    InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Learning multi-scale photo exposure correction. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 9157–9167

    work page

  3. [3]

    Jong-Hyeon Baek, DaeHyun Kim, Su-Min Choi, Hyo-jun Lee, Hanul Kim, and Yeong Jun Koh. 2023. Luminance-aware color transform for multiple exposure correction. InProceedings of the IEEE/CVF international conference on computer vision. 6156–6165

    work page 2023

  4. [4]

    Peter L Bartlett, Dylan J Foster, and Matus J Telgarsky. 2017. Spectrally- normalized margin bounds for neural networks.Advances in neural information processing systems30 (2017)

    work page 2017

  5. [5]

    Vladimir Bychkovsky, Sylvain Paris, Eric Chan, and Frédo Durand. 2011. Learning photographic global tonal adjustment with a database of input/output image pairs. InCVPR 2011. IEEE, 97–104

    work page 2011

  6. [6]

    Jianrui Cai, Shuhang Gu, and Lei Zhang. 2018. Learning a deep single image contrast enhancer from multi-exposure images.IEEE transactions on image processing27, 4 (2018), 2049–2062

    work page 2018

  7. [7]

    Chris Careaga and Yağız Aksoy. 2024. Colorful diffuse intrinsic image decompo- sition in the wild.ACM Transactions on Graphics (TOG)43, 6 (2024), 1–12

    work page 2024

  8. [8]

    Chen Chen, Qifeng Chen, Jia Xu, and Vladlen Koltun. 2018. Learning to see in the dark. InProceedings of the IEEE conference on computer vision and pattern recognition. 3291–3300

    work page 2018

  9. [9]

    Sixiang Chen, Tian Ye, Jinbin Bai, Erkang Chen, Jun Shi, and Lei Zhu. 2023. Sparse sampling transformer with uncertainty-driven ranking for unified removal of raindrops and rain streaks. InProceedings of the IEEE/CVF international conference on computer vision. 13106–13117

    work page 2023

  10. [10]

    Sixiang Chen, Tian Ye, Chenghao Xue, Haoyu Chen, Yun Liu, Erkang Chen, and Lei Zhu. 2023. Uncertainty-driven dynamic degradation perceiving and background modeling for efficient single image desnowing. InProceedings of the 31st ACM international conference on multimedia. 4269–4280

    work page 2023

  11. [11]

    Ruodai Cui, Li Niu, and Guosheng Hu. 2024. Unsupervised Exposure Correction. InEuropean Conference on Computer Vision. Springer, 252–268

    work page 2024

  12. [13]

    Zhenxuan Fang, Weisheng Dong, Xin Li, Jinjian Wu, Leida Li, and Guangming Shi. 2022. Uncertainty learning in kernel estimation for multi-stage blind image super-resolution. InEuropean conference on computer vision. Springer, 144–161

    work page 2022

  13. [14]

    Mário Figueiredo. 2001. Adaptive sparseness using Jeffreys prior.Advances in neural information processing systems14 (2001)

    work page 2001

  14. [15]

    Zhenqi Fu, Yan Yang, Xiaotong Tu, Yue Huang, Xinghao Ding, and Kai-Kuang Ma. 2023. Learning a simple low-light image enhancer from paired low-light instances. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition. 22252–22261

    work page 2023

  15. [16]

    Chunle Guo, Chongyi Li, Jichang Guo, Chen Change Loy, Junhui Hou, Sam Kwong, and Runmin Cong. 2020. Zero-reference deep curve estimation for low- light image enhancement. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition. 1780–1789

    work page 2020

  16. [17]

    Jie Huang, Yajing Liu, Xueyang Fu, Man Zhou, Yang Wang, Feng Zhao, and Zhiwei Xiong. 2022. Exposure normalization and compensation for multiple-exposure correction. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 6043–6052

    work page 2022

  17. [18]

    Jie Huang, Yajing Liu, Feng Zhao, Keyu Yan, Jinghao Zhang, Yukun Huang, Man Zhou, and Zhiwei Xiong. 2022. Deep fourier-based exposure correction network with spatial-frequency interaction. InEuropean Conference on Computer Vision. Springer, 163–180

    work page 2022

  18. [19]

    Jie Huang, Feng Zhao, Man Zhou, Jie Xiao, Naishan Zheng, Kaiwen Zheng, and Zhiwei Xiong. 2023. Learning sample relationship for exposure correction. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 9904–9913

    work page 2023

  19. [20]

    Jie Huang, Man Zhou, Yajing Liu, Mingde Yao, Feng Zhao, and Zhiwei Xiong

    work page

  20. [21]

    In Proceedings of the 30th ACM International Conference on Multimedia

    Exposure-consistency representation learning for exposure correction. In Proceedings of the 30th ACM International Conference on Multimedia. 6309–6317

    work page

  21. [22]

    Adam: A Method for Stochastic Optimization

    Diederik P. Kingma and Jimmy Ba. 2017. Adam: A Method for Stochastic Opti- mization. arXiv:1412.6980 [cs.LG] https://arxiv.org/abs/1412.6980

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  22. [23]

    Chongyi Li, Chunle Guo, and Chen Change Loy. 2021. Learning to enhance low-light image via zero-reference deep curve estimation.IEEE transactions on pattern analysis and machine intelligence44, 8 (2021), 4225–4238

    work page 2021

  23. [24]

    Gehui Li, Bin Chen, Chen Zhao, Lei Zhang, and Jian Zhang. 2025. Osmamba: Omnidirectional spectral mamba with dual-domain prior generator for expo- sure correction. InProceedings of the Computer Vision and Pattern Recognition Conference. 7480–7490

    work page 2025

  24. [25]

    Gehui Li, Jinyuan Liu, Long Ma, Zhiying Jiang, Xin Fan, and Risheng Liu. 2023. Fearless luminance adaptation: A macro-micro-hierarchical transformer for ex- posure correction. InProceedings of the 31st ACM International Conference on Multimedia. 7304–7313

    work page 2023

  25. [26]

    Yiyu Li, Ke Xu, Gerhard Petrus Hancke, and Rynson W.H. Lau. 2024. Color Shift Estimation-and-Correction for Image Enhancement. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 25389– 25398

    work page 2024

  26. [27]

    Ziwen Li, Feng Zhang, Meng Cao, Jinpu Zhang, Yuanjie Shao, Yuehuan Wang, and Nong Sang. 2024. Real-time exposure correction via collaborative transformations and adaptive sampling. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2984–2994

    work page 2024

  27. [28]

    Dong Liang, Zhengyan Xu, Ling Li, Mingqiang Wei, and Songcan Chen. 2024. Pie: Physics-inspired low-light enhancement.International Journal of Computer Vision132, 9 (2024), 3911–3932

    work page 2024

  28. [29]

    Kyungsun Lim, Nyeong-Ho Shin, Young-Yoon Lee, and Chang-Su Kim. 2020. Order Learning and Its Application to Age Estimation. InInternational Conference on Learning Representations

    work page 2020

  29. [30]

    Jin Liu, Huiyuan Fu, Chuanming Wang, and Huadong Ma. 2024. Region-aware exposure consistency network for mixed exposure correction. InProceedings of the AAAI Conference on Artificial Intelligence, Vol. 38. 3648–3656

    work page 2024

  30. [31]

    Risheng Liu, Long Ma, Jiaao Zhang, Xin Fan, and Zhongxuan Luo. 2021. Retinex- inspired unrolling with cooperative prior architecture search for low-light image enhancement. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition. 10561–10570

    work page 2021

  31. [32]

    Long Ma, Tengyu Ma, Risheng Liu, Xin Fan, and Zhongxuan Luo. 2022. Toward fast, flexible, and robust low-light image enhancement. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition. 5637–5646

    work page 2022

  32. [33]

    Tom Mertens, Jan Kautz, and Frank Van Reeth. 2009. Exposure fusion: A sim- ple and practical alternative to high dynamic range photography. InComputer graphics forum, Vol. 28. Wiley Online Library, 161–171

    work page 2009

  33. [34]

    completely blind

    Anish Mittal, Rajiv Soundararajan, and Alan C Bovik. 2012. Making a “completely blind” image quality analyzer.IEEE Signal processing letters20, 3 (2012), 209–212

    work page 2012

  34. [35]

    Qian Ning, Weisheng Dong, Xin Li, Jinjian Wu, and Guangming Shi. 2021. Uncertainty-driven loss for single image super-resolution.Advances in Neu- ral Information Processing Systems34 (2021), 16398–16409

    work page 2021

  35. [36]

    Yusuke Tsuzuku, Issei Sato, and Masashi Sugiyama. 2018. Lipschitz-margin train- ing: Scalable certification of perturbation invariance for deep neural networks. Advances in neural information processing systems31 (2018)

    work page 2018

  36. [37]

    Bo Wang, Huiyuan Fu, Zhiye Huang, Siru Zhang, Xin Wang, and Huadong Ma

    work page

  37. [38]

    InProceedings of the IEEE/CVF International Conference on Computer Vision

    From Abyssal Darkness to Blinding Glare: A Benchmark on Extreme Exposure Correction in Real World. InProceedings of the IEEE/CVF International Conference on Computer Vision. 7666–7675

    work page

  38. [39]

    Bokang Wang, Qian Ning, Fangfang Wu, Xin Li, Weisheng Dong, and Guang- ming Shi. 2024. Uncertainty modeling of the transmission map for single image dehazing.IEEE Transactions on Circuits and Systems for Video Technology34, 11 (2024), 11115–11127

    work page 2024

  39. [40]

    Haoyuan Wang, Ke Xu, and Rynson WH Lau. 2022. Local color distributions prior for image enhancement. InEuropean conference on computer vision. Springer, 343–359

    work page 2022

  40. [41]

    Jiayun Wang, Yubei Chen, Rudrasis Chakraborty, and Stella X. Yu. 2020. Orthog- onal Convolutional Neural Networks. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    work page 2020

  41. [42]

    Zhou Wang, Alan C Bovik, Hamid R Sheikh, and Eero P Simoncelli. 2004. Image quality assessment: from error visibility to structural similarity.IEEE transactions on image processing13, 4 (2004), 600–612

    work page 2004

  42. [43]

    Chen Wei, Wenjing Wang, Wenhan Yang, and Jiaying Liu. 2018. Deep Retinex Decomposition for Low-Light Enhancement. InBritish Machine Vision Confer- ence

    work page 2018

  43. [44]

    Xin Xu, Hao Liu, Wei Liu, Wei Wang, Jiayi Wu, and Kui Jiang. 2025. ICLR: Inter-Chrominance and Luminance Interaction for Natural Color Restoration in Low-Light Image Enhancement. arXiv:2511.13607 [cs.CV] https://arxiv.org/abs/ 2511.13607

    work page arXiv 2025

  44. [45]

    Qingsen Yan, Yixu Feng, Cheng Zhang, Guansong Pang, Kangbiao Shi, Peng Wu, Wei Dong, Jinqiu Sun, and Yanning Zhang. 2025. Hvi: A new color space for low-light image enhancement. InProceedings of the computer vision and pattern recognition conference. 5678–5687. , ,

    work page 2025

  45. [46]

    Qingsen Yan, Kangbiao Shi, Yixu Feng, Tao Hu, Peng Wu, Guansong Pang, and Yanning Zhang. 2025. HVI-CIDNet+: Beyond Extreme Darkness for Low-Light Image Enhancement. arXiv:2507.06814 [cs.CV] https://arxiv.org/abs/2507.06814

    work page arXiv 2025

  46. [47]

    Canqian Yang, Meiguang Jin, Xu Jia, Yi Xu, and Ying Chen. 2022. AdaInt: Learning adaptive intervals for 3D lookup tables on real-time image enhancement. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 17522–17531

    work page 2022

  47. [48]

    Wenhan Yang, Shiqi Wang, Yuming Fang, Yue Wang, and Jiaying Liu. 2020. From fidelity to perceptual quality: A semi-supervised approach for low-light image enhancement. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition. 3063–3072

    work page 2020

  48. [49]

    Syed Waqas Zamir, Aditya Arora, Salman Khan, Munawar Hayat, Fahad Shah- baz Khan, and Ming-Hsuan Yang. 2022. Restormer: Efficient transformer for high-resolution image restoration. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition. 5728–5739

    work page 2022

  49. [50]

    Hui Zeng, Jianrui Cai, Lida Li, Zisheng Cao, and Lei Zhang. 2020. Learning image-adaptive 3d lookup tables for high performance photo enhancement in real-time.IEEE Transactions on Pattern Analysis and Machine Intelligence44, 4 (2020), 2058–2073

    work page 2020

  50. [51]

    Richard Zhang, Phillip Isola, Alexei A Efros, Eli Shechtman, and Oliver Wang

    work page

  51. [52]

    InProceedings of the IEEE conference on computer vision and pattern recognition

    The unreasonable effectiveness of deep features as a perceptual metric. InProceedings of the IEEE conference on computer vision and pattern recognition. 586–595

    work page