pith. machine review for the scientific record. sign in

arxiv: 2605.02464 · v1 · submitted 2026-05-04 · 💻 cs.CV

Recognition: 2 theorem links

ExpoCM: Exposure-Aware One-Step Generative Single-Image HDR Reconstruction

Aoyu Liu , Zhen Liu , Ziyi Wang , Dian Chen , Bing Zeng , Shuaicheng Liu
Authors on Pith no claims yet

Pith reviewed 2026-05-08 18:29 UTC · model grok-4.3

classification 💻 cs.CV
keywords single-image HDR reconstructionone-step generative modelexposure-awareconsistency trajectoriesprobability flow ODEsoft exposure maskperceptual image quality
0
0 comments X

The pith

ExpoCM performs single-image HDR reconstruction in one inference step by building exposure-aware consistency trajectories inside a probability flow ODE.

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

Single-image HDR reconstruction must restore lost detail in saturated bright areas and reduce noise in dark areas, yet most diffusion methods require many sampling steps and treat every pixel the same regardless of local exposure. The paper shows that a soft mask can first divide the input into over-exposed, under-exposed, and well-exposed regions, after which region-specific perturbations produce tailored consistency trajectories that let the model hallucinate plausible content and suppress noise in a single forward pass. An additional exposure-guided loss in CIE L*a*b* space further balances brightness and color. The resulting model matches or exceeds the fidelity of prior iterative approaches on three public benchmarks while running hundreds of times faster.

Core claim

By reformulating HDR reconstruction as a probability flow ODE and constructing exposure-dependent perturbations that follow region-conditioned consistency trajectories, the method recovers high-dynamic-range radiance from a single LDR image in one distillation-free step.

What carries the argument

Exposure-aware consistency trajectories constructed from exposure-dependent perturbations inside a probability flow ODE, conditioned by a soft exposure mask that partitions the image into over-, under-, and well-exposed regions.

If this is right

  • The method reaches state-of-the-art fidelity and perceptual accuracy on the HDR-REAL, HDR-EYE, and AIM2025 benchmarks.
  • Inference is more than 400 times faster than a 1000-step DDPM and more than 20 times faster than a 50-step DDIM.
  • The exposure-guided luminance-chromaticity loss reduces brightness bias and color drift compared with uniform losses.
  • Region-conditioned generation preserves reliable structures in well-exposed areas while selectively enhancing the other two regions.

Where Pith is reading between the lines

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

  • The same mask-and-trajectory idea could extend to other spatially varying degradations such as non-uniform blur or sensor noise.
  • One-step inference removes the main obstacle to running HDR reconstruction on mobile or embedded cameras in real time.
  • Because the trajectories are derived from the ODE rather than from a fixed schedule, the framework may generalize to video sequences by adding temporal consistency constraints on the mask.

Load-bearing premise

A soft exposure mask can reliably partition the image so that the resulting region-conditioned trajectories hallucinate plausible details and suppress noise without introducing visible artifacts or inconsistencies at region boundaries.

What would settle it

Reconstructed HDR images on test photographs that contain adjacent over-exposed and under-exposed regions show visible seams, color shifts, or hallucinated textures exactly along the mask boundaries.

Figures

Figures reproduced from arXiv: 2605.02464 by Aoyu Liu, Bing Zeng, Dian Chen, Shuaicheng Liu, Zhen Liu, Ziyi Wang.

Figure 1
Figure 1. Figure 1: Visual comparisons with previous state-of-the-art meth view at source ↗
Figure 2
Figure 2. Figure 2: The overall pipeline of our proposed ExpoCM framework. The exposure mask generation module first partitions the input LDR view at source ↗
Figure 3
Figure 3. Figure 3: Qualitative comparisons with state-of-the-art single-image HDR reconstruction methods on the AIM2025 and HDR-REAL view at source ↗
Figure 5
Figure 5. Figure 5: Visual comparisons of our ablation studies on the pro view at source ↗
read the original abstract

Single-image HDR reconstruction aims to recover high dynamic range radiance from a single low dynamic range (LDR) input, but remains highly ill-posed due to detail saturation in over-exposed regions and noise amplification in under-exposed areas. While recent diffusion-based approaches offer powerful generative priors, they often overlook the exposure-dependent nature of the degradation and incur substantial computational costs from iterative sampling. To address these challenges, we propose ExpoCM, a novel one-step generative HDR reconstruction framework that reformulates HDR reconstruction as a Probability Flow ODE (PF-ODE) and constructs exposure-aware consistency trajectories via exposure-dependent perturbations. Specifically, a soft exposure mask is first constructed to separate the LDR image into over-, under-, and well-exposed regions. Based on this partition, region-conditioned consistency trajectories are designed to hallucinate saturated details, suppress noise in dark regions, and preserve reliable structures within a single, distillation-free inference step. To further enhance perceptual quality, we introduce an Exposure-guided Luminance-Chromaticity Loss in the CIE~$\text{L}^*\text{a}^*\text{b}^*$ space, which assigns exposure-aware weights to luminance and chromaticity components, effectively mitigating brightness bias and color drift. Extensive experiments on the HDR-REAL, HDR-EYE, and AIM2025 benchmarks demonstrate that ExpoCM achieves state-of-the-art fidelity and perceptual accuracy, while enabling over 400$\times$ and 20$\times$ faster inference compared to DDPM (1000 steps) and DDIM (50 steps), respectively.

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 ExpoCM, a one-step generative framework for single-image HDR reconstruction. It reformulates the problem using the probability flow ODE and introduces a soft exposure mask to partition the LDR input into over-, under-, and well-exposed regions, from which region-conditioned consistency trajectories are derived to hallucinate details and suppress noise in a single distillation-free step. An Exposure-guided Luminance-Chromaticity Loss in CIE L*a*b* space is added to mitigate brightness and color biases. Experiments on HDR-REAL, HDR-EYE, and AIM2025 benchmarks are reported to show SOTA fidelity/perceptual quality together with >400× and >20× speedups versus DDPM (1000 steps) and DDIM (50 steps).

Significance. If the performance and boundary-consistency claims hold after proper verification, the work would be significant for practical HDR imaging: it demonstrates that exposure-aware partitioning can be combined with a single PF-ODE step to achieve both quality and extreme efficiency gains over iterative diffusion baselines. The exposure-dependent trajectory design and Lab-space loss constitute a concrete technical contribution that directly targets the ill-posedness of saturation and noise in LDR-to-HDR mapping.

major comments (3)
  1. [Abstract and §3] Abstract and §3 (method description): no equations or algorithmic details are supplied for constructing the soft exposure mask, for defining the exposure-dependent perturbations, or for blending the region-conditioned PF-ODE trajectories. Without these, it is impossible to assess whether boundary continuity is enforced or whether the claimed absence of seams and color shifts is actually achieved.
  2. [§4] §4 (experiments): the SOTA claims on HDR-REAL, HDR-EYE, and AIM2025 rest on aggregate metrics with no reported training protocol, baseline re-implementation details, statistical significance tests, or ablation studies isolating the mask and trajectory components. This renders the central fidelity and speedup assertions unverifiable from the manuscript.
  3. [§3.2 and §4.3] §3.2 and §4.3: the weakest assumption—that a soft mask plus single-step region conditioning produces seamless, artifact-free outputs—is not supported by any boundary-specific analysis, visual insets, or quantitative seam metrics. This directly undermines the one-step generative claim.
minor comments (2)
  1. [§3] Notation for the PF-ODE and consistency trajectories should be introduced with explicit references to the underlying diffusion literature to avoid ambiguity.
  2. [Figures 3–5] Figure captions and axis labels in the qualitative results could be expanded to indicate exposure-region boundaries and highlight any residual artifacts.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback. We address each major comment point by point below and will make substantial revisions to improve clarity, verifiability, and supporting evidence in the manuscript.

read point-by-point responses

  1. Referee: [Abstract and §3] Abstract and §3 (method description): no equations or algorithmic details are supplied for constructing the soft exposure mask, for defining the exposure-dependent perturbations, or for blending the region-conditioned PF-ODE trajectories. Without these, it is impossible to assess whether boundary continuity is enforced or whether the claimed absence of seams and color shifts is actually achieved.

    Authors: We agree that explicit mathematical formulations are necessary for full reproducibility and assessment. In the revised manuscript, we will add the precise equations in §3 for (i) soft exposure mask construction via per-pixel logistic exposure estimation, (ii) exposure-dependent perturbation schedules applied to the PF-ODE, and (iii) the soft-blending operator that combines region-conditioned consistency trajectories. These additions will explicitly show how soft weighting enforces boundary continuity and suppresses seams/color shifts. The current textual description outlines the high-level design, but we acknowledge the need for formal details. revision: yes

  2. Referee: [§4] §4 (experiments): the SOTA claims on HDR-REAL, HDR-EYE, and AIM2025 rest on aggregate metrics with no reported training protocol, baseline re-implementation details, statistical significance tests, or ablation studies isolating the mask and trajectory components. This renders the central fidelity and speedup assertions unverifiable from the manuscript.

    Authors: We accept that greater experimental transparency is required. The revised §4 will include: full training protocol (hyperparameters, optimizer, learning rate schedule, and data splits); exact re-implementation details for DDPM (1000 steps) and DDIM (50 steps) baselines using their official repositories with our adaptations; statistical significance testing (paired t-tests on PSNR, SSIM, and LPIPS across the test sets with p-values); and ablation studies that isolate the soft exposure mask and region-conditioned trajectories. These changes will render the fidelity and speedup claims verifiable. revision: yes

  3. Referee: [§3.2 and §4.3] §3.2 and §4.3: the weakest assumption—that a soft mask plus single-step region conditioning produces seamless, artifact-free outputs—is not supported by any boundary-specific analysis, visual insets, or quantitative seam metrics. This directly undermines the one-step generative claim.

    Authors: We recognize that targeted evidence for boundary behavior is currently insufficient. In the revision, we will augment §4.3 with boundary-specific analysis: zoomed visual insets centered on exposure transition regions, and new quantitative metrics including boundary gradient consistency error and a perceptual seam artifact score. These will be reported for ExpoCM versus baselines to directly support the seam-free, artifact-free claim of the single-step approach. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected in derivation chain

full rationale

The paper's core derivation reformulates HDR reconstruction as a standard PF-ODE and introduces novel elements (soft exposure mask partitioning, region-conditioned consistency trajectories via exposure-dependent perturbations, and an Exposure-guided Luminance-Chromaticity Loss in CIE L*a*b* space) as explicit constructions rather than reductions to fitted inputs or prior results by definition. No equations or claims reduce by construction to self-referential quantities, and performance assertions rest on empirical evaluation against external benchmarks rather than internal equivalence. The framework draws on established diffusion literature for PF-ODE without load-bearing self-citation chains or ansatz smuggling.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The method rests on standard mathematical properties of probability flow ODEs from the diffusion modeling literature and domain assumptions about exposure-dependent image degradation; no new physical entities or ad-hoc axioms are introduced.

axioms (1)
  • standard math Probability flow ODEs can be used to construct deterministic generative trajectories from diffusion models
    Invoked when reformulating HDR reconstruction as a PF-ODE for one-step sampling.

pith-pipeline@v0.9.0 · 5581 in / 1248 out tokens · 78633 ms · 2026-05-08T18:29:54.970118+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

58 extracted references · 4 canonical work pages · 1 internal anchor

  1. [1]

    Pu21: A novel perceptually uniform encoding for adapting existing quality metrics for hdr

    Maryam Azimi et al. Pu21: A novel perceptually uniform encoding for adapting existing quality metrics for hdr. In 2021 Picture Coding Symposium (PCS), pages 1–5. IEEE,

    2021

  2. [2]

    Inverse tone mapping

    Francesco Banterle, Patrick Ledda, Kurt Debattista, and Alan Chalmers. Inverse tone mapping. InCGIT, pages 349– 356, 2006. 2

    2006

  3. [3]

    A framework for inverse tone mapping.The Visual Computer, 23(7):467–478, 2007

    Francesco Banterle, Patrick Ledda, Kurt Debattista, Alan Chalmers, and Marina Bloj. A framework for inverse tone mapping.The Visual Computer, 23(7):467–478, 2007. 2

    2007

  4. [4]

    High dynamic range imaging and low dy- namic range expansion for generating hdr content

    Francesco Banterle, Kurt Debattista, Alessandro Artusi, Sumanta Pattanaik, Karol Myszkowski, Patrick Ledda, and Alan Chalmers. High dynamic range imaging and low dy- namic range expansion for generating hdr content. InCom- put. Graph. Forum, pages 2343–2367. Wiley Online Library,

  5. [5]

    Improv- ing dynamic hdr imaging with fusion transformer

    Rufeng Chen, Bolun Zheng, Hua Zhang, Quan Chen, Cheng- gang Yan, Gregory Slabaugh, and Shanxin Yuan. Improv- ing dynamic hdr imaging with fusion transformer. InAAAI, pages 340–349, 2023. 2

    2023

  6. [6]

    Hdrunet: Single image hdr reconstruction with denoising and dequantization

    Xiangyu Chen, Yihao Liu, Zhengwen Zhang, Yu Qiao, and Chao Dong. Hdrunet: Single image hdr reconstruction with denoising and dequantization. InCVPR, pages 354–363,

  7. [7]

    Blind-spot guided diffusion for self-supervised real-world denoising.arXiv preprint arXiv:2509.16091, 2025

    Shen Cheng, Haipeng Li, Haibin Huang, Xiaohong Liu, and Shuaicheng Liu. Blind-spot guided diffusion for self-supervised real-world denoising.arXiv preprint arXiv:2509.16091, 2025. 3

    work page arXiv 2025

  8. [8]

    Single image ldr to hdr conversion us- ing conditional diffusion

    Dwip Dalal, Gautam Vashishtha, Prajwal Singh, and Shan- muganathan Raman. Single image ldr to hdr conversion us- ing conditional diffusion. InICIP, pages 3533–3537. IEEE,

  9. [9]

    Hdr image reconstruction from a single exposure using deep cnns.ACM TOG, 36(6):1–15,

    Gabriel Eilertsen, Joel Kronander, Gyorgy Denes, Rafał K Mantiuk, and Jonas Unger. Hdr image reconstruction from a single exposure using deep cnns.ACM TOG, 36(6):1–15,

  10. [10]

    Burst photography for high dynamic range and low-light imaging on mobile cameras.ACM TOG, 35(6):1– 12, 2016

    Samuel W Hasinoff, Dillon Sharlet, Ryan Geiss, Andrew Adams, Jonathan T Barron, Florian Kainz, Jiawen Chen, and Marc Levoy. Burst photography for high dynamic range and low-light imaging on mobile cameras.ACM TOG, 35(6):1– 12, 2016. 2

    2016

  11. [11]

    Reti-diff: Illumination degradation image restoration with retinex-based latent diffusion model

    Chunming He, Chengyu Fang, Yulun Zhang, Kai Li, Longx- iang Tang, Chenyu You, Fengyang Xiao, Zhenhua Guo, and Xiu Li. Reti-diff: Illumination degradation image restoration with retinex-based latent diffusion model. InICLR, 2025. 6

    2025

  12. [12]

    Denoising dif- fusion probabilistic models

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising dif- fusion probabilistic models. InNeurIPS, pages 6840–6851,

  13. [13]

    Hdr deghosting: How to deal with saturation? InCVPR, pages 1163–1170, 2013

    Jun Hu, Orazio Gallo, Kari Pulli, and Xiaobai Sun. Hdr deghosting: How to deal with saturation? InCVPR, pages 1163–1170, 2013. 2

    2013

  14. [14]

    Detail-preserving diffusion mod- els for low-light image enhancement.IEEE TCSVT, 2024

    Yan Huang, Xiaoshan Liao, Jinxiu Liang, Boxin Shi, Yong Xu, and Patrick Le Callet. Detail-preserving diffusion mod- els for low-light image enhancement.IEEE TCSVT, 2024. 3

    2024

  15. [15]

    Physiological inverse tone mapping based on retina re- sponse.The Visual Computer, 30(5):507–517, 2014

    Yongqing Huo, Fan Yang, Le Dong, and Vincent Brost. Physiological inverse tone mapping based on retina re- sponse.The Visual Computer, 30(5):507–517, 2014. 2

    2014

  16. [16]

    Low-light image enhancement with wavelet-based diffusion models.ACM TOG, 42(6):1–14,

    Hai Jiang, Ao Luo, Songchen Han, Haoqiang Fan, and Shuaicheng Liu. Low-light image enhancement with wavelet-based diffusion models.ACM TOG, 42(6):1–14,

  17. [17]

    Deep high dynamic range imaging of dynamic scenes.ACM TOG, 36 (4):144, 2017

    Nima Khademi Kalantari and Ravi Ramamoorthi. Deep high dynamic range imaging of dynamic scenes.ACM TOG, 36 (4):144, 2017. 1, 2

    2017

  18. [18]

    Denoising diffusion restoration models

    Bahjat Kawar, Michael Elad, Stefano Ermon, and Jiaming Song. Denoising diffusion restoration models. InNeurIPS, pages 23593–23606, 2022. 3

    2022

  19. [19]

    Imagic: Text-based real image editing with diffusion models

    Bahjat Kawar, Shiran Zada, Oran Lang, Omer Tov, Huiwen Chang, Tali Dekel, Inbar Mosseri, and Michal Irani. Imagic: Text-based real image editing with diffusion models. In CVPR, pages 6007–6017, 2023. 3

    2023

  20. [20]

    Ghost removal in high dynamic range images

    Erum Arif Khan, Ahmet Oguz Akyuz, and Erik Reinhard. Ghost removal in high dynamic range images. InICIP, pages 2005–2008, 2006. 2

    2005

  21. [21]

    Deep re- cursive hdri: Inverse tone mapping using generative adver- sarial networks

    Siyeong Lee, Gwon Hwan An, and Suk-Ju Kang. Deep re- cursive hdri: Inverse tone mapping using generative adver- sarial networks. InECCV, pages 596–611, 2018. 2, 3

    2018

  22. [22]

    Exposure-limited image enhancement with generative diffu- sion prior

    Baiang Li, Sizhuo Ma, Yanhong Zeng, Xiaogang Xu, Youqing Fang, Zhao Zhang, Jian Wang, and Kai Chen. Exposure-limited image enhancement with generative diffu- sion prior. In2025 IEEE International Conference on Com- putational Photography (ICCP), pages 1–10. IEEE, 2025. 3, 5

    2025

  23. [23]

    Dmhomo: Learning ho- mography with diffusion models.ACM TOG, 43(3):1–16,

    Haipeng Li, Hai Jiang, Ao Luo, Ping Tan, Haoqiang Fan, Bing Zeng, and Shuaicheng Liu. Dmhomo: Learning ho- mography with diffusion models.ACM TOG, 43(3):1–16,

  24. [24]

    Single-image hdr reconstruction by learning to reverse the camera pipeline

    Yu-Lun Liu, Wei-Sheng Lai, Yu-Sheng Chen, Yi-Lung Kao, Ming-Hsuan Yang, Yung-Yu Chuang, and Jia-Bin Huang. Single-image hdr reconstruction by learning to reverse the camera pipeline. InCVPR, pages 1651–1660, 2020. 2, 3, 5, 6, 8

    2020

  25. [25]

    Adnet: Attention-guided deformable convolutional network for high dynamic range imaging

    Zhen Liu, Wenjie Lin, Xinpeng Li, Qing Rao, Ting Jiang, Mingyan Han, Haoqiang Fan, Jian Sun, and Shuaicheng Liu. Adnet: Attention-guided deformable convolutional network for high dynamic range imaging. InCVPRW, pages 463– 470, 2021. 1, 2

    2021

  26. [26]

    Ghost-free high dynamic range imaging with context-aware transformer

    Zhen Liu, Yinglong Wang, Bing Zeng, and Shuaicheng Liu. Ghost-free high dynamic range imaging with context-aware transformer. InECCV, pages 344–360. Springer, 2022. 1, 2, 6

    2022

  27. [27]

    Solving ill-posed regions in high dynamic range re- construction with uncertainty-aware diffusion models.IEEE TCSVT, 2025

    Zhen Liu, Hai Jiang, Haipeng Li, Shuaicheng Liu, and Bing Zeng. Solving ill-posed regions in high dynamic range re- construction with uncertainty-aware diffusion models.IEEE TCSVT, 2025. 5

    2025

  28. [28]

    Raw-flow: Advancing rgb-to-raw image reconstruction with deterministic latent flow matching.AAAI, 40(9):7431– 7439, 2026

    Zhen Liu, Diedong Feng, Hai Jiang, Liaoyuan Zeng, Hao Wang, Chaoyu Feng, Lei Lei, Bing Zeng, and Shuaicheng Liu. Raw-flow: Advancing rgb-to-raw image reconstruction with deterministic latent flow matching.AAAI, 40(9):7431– 7439, 2026. 3

    2026

  29. [29]

    Repaint: Inpainting using denoising diffusion probabilistic models

    Andreas Lugmayr, Martin Danelljan, Andres Romero, Fisher Yu, Radu Timofte, and Luc Van Gool. Repaint: Inpainting using denoising diffusion probabilistic models. InCVPR, pages 11461–11471, 2022. 3

    2022

  30. [30]

    Learning effi- cient meshflow and optical flow from event cameras.IEEE TPAMI, 48(2):1355–1372, 2026

    Xinglong Luo, Ao Luo, Kunming Luo, Zhengning Wang, Ping Tan, Bing Zeng, and Shuaicheng Liu. Learning effi- cient meshflow and optical flow from event cameras.IEEE TPAMI, 48(2):1355–1372, 2026. 2

    2026

  31. [31]

    Deep guided learning for fast multi- exposure image fusion.IEEE TIP, 29:2808–2819, 2019

    Kede Ma, Zhengfang Duanmu, Hanwei Zhu, Yuming Fang, and Zhou Wang. Deep guided learning for fast multi- exposure image fusion.IEEE TIP, 29:2808–2819, 2019. 2

    2019

  32. [32]

    Hdr-vdp-2: A calibrated visual metric for visibility and quality predictions in all luminance conditions

    Rafał Mantiuk, Kil Joong Kim, Allan G Rempel, and Wolf- gang Heidrich. Hdr-vdp-2: A calibrated visual metric for visibility and quality predictions in all luminance conditions. ACM TOG, 30(4):1–14, 2011. 6

    2011

  33. [33]

    2304.13625 , archivePrefix=

    Rafal K Mantiuk, Dounia Hammou, and Param Hanji. Hdr- vdp-3: A multi-metric for predicting image differences, quality and contrast distortions in high dynamic range and regular content.arXiv preprint arXiv:2304.13625, 2023. 6

    work page arXiv 2023

  34. [34]

    Expandnet: A deep convo- lutional neural network for high dynamic range expansion from low dynamic range content

    Demetris Marnerides, Thomas Bashford-Rogers, Jonathan Hatchett, and Kurt Debattista. Expandnet: A deep convo- lutional neural network for high dynamic range expansion from low dynamic range content. InComput. Graph. Forum, pages 37–49. Wiley Online Library, 2018. 2, 3, 6

    2018

  35. [35]

    Exposure fusion

    Tom Mertens, Jan Kautz, and Frank Van Reeth. Exposure fusion. InPG, pages 382–390, 2007. 1, 2

    2007

  36. [36]

    Visual attention in ldr and hdr images

    Hiromi Nemoto, Pavel Korshunov, Philippe Hanhart, and Touradj Ebrahimi. Visual attention in ldr and hdr images. InVPQM, page 6, 2015. 5, 6

    2015

  37. [37]

    Elucidating the exposure bias in diffusion models

    Mang Ning, Mingxiao Li, Jianlin Su, Albert Ali Salah, and Itir Onal Ertugrul. Elucidating the exposure bias in diffusion models. InICLR, 2024. 2

    2024

  38. [38]

    Hdr-gan: Hdr image reconstruction from multi-exposed ldr images with large motions.IEEE TIP, 30: 3885–3896, 2021

    Yuzhen Niu, Jianbin Wu, Wenxi Liu, Wenzhong Guo, and Rynson WH Lau. Hdr-gan: Hdr image reconstruction from multi-exposed ldr images with large motions.IEEE TIP, 30: 3885–3896, 2021. 2

    2021

  39. [39]

    Robust high dynamic range imaging by rank mini- mization.IEEE TPAMI, 37(6):1219–1232, 2014

    Tae-Hyun Oh, Joon-Young Lee, Yu-Wing Tai, and In So Kweon. Robust high dynamic range imaging by rank mini- mization.IEEE TPAMI, 37(6):1219–1232, 2014. 2

    2014

  40. [40]

    Labeled from unlabeled: Exploiting unlabeled data for few-shot deep hdr deghosting

    K Ram Prabhakar, Gowtham Senthil, Susmit Agrawal, R Venkatesh Babu, and Rama Krishna Sai S Gorthi. Labeled from unlabeled: Exploiting unlabeled data for few-shot deep hdr deghosting. InCVPR, pages 4875–4885, 2021. 2

    2021

  41. [41]

    Multiscale structure guided diffusion for image deblurring

    Mengwei Ren, Mauricio Delbracio, Hossein Talebi, Guido Gerig, and Peyman Milanfar. Multiscale structure guided diffusion for image deblurring. InICCV, pages 10721– 10733, 2023. 3

    2023

  42. [42]

    U-net: Convolutional networks for biomedical image segmentation

    Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-net: Convolutional networks for biomedical image segmentation. InMICCAI, pages 234–241, 2015. 5

    2015

  43. [43]

    Single image hdr reconstruction using a cnn with masked features and perceptual loss.ACM TOG, 2020

    Marcel Santana Santos, Tsang Ing Ren, and Nima Khademi Kalantari. Single image hdr reconstruction using a cnn with masked features and perceptual loss.ACM TOG, 2020. 2

    2020

  44. [44]

    Ro- bust patch-based hdr reconstruction of dynamic scenes.ACM TOG, 31(6):203, 2012

    Pradeep Sen, Nima Khademi Kalantari, Maziar Yaesoubi, Soheil Darabi, Dan B Goldman, and Eli Shechtman. Ro- bust patch-based hdr reconstruction of dynamic scenes.ACM TOG, 31(6):203, 2012. 2

    2012

  45. [45]

    Denois- ing diffusion implicit models

    Jiaming Song, Chenlin Meng, and Stefano Ermon. Denois- ing diffusion implicit models. InICLR, 2021. 3, 6

    2021

  46. [46]

    Score-Based Generative Modeling through Stochastic Differential Equations

    Yang Song, Jascha Sohl-Dickstein, Diederik P Kingma, Ab- hishek Kumar, Stefano Ermon, and Ben Poole. Score-based generative modeling through stochastic differential equa- tions.arXiv preprint arXiv:2011.13456, 2020. 3

    work page Pith review arXiv 2011

  47. [47]

    Consistency Models

    Yang Song, Prafulla Dhariwal, Mark Chen, and Ilya Sutskever. Consistency models.arXiv preprint arXiv:2303.01469, 2023. 3

    work page internal anchor Pith review arXiv 2023

  48. [48]

    Alignment-free hdr deghosting with semantics con- sistent transformer

    Steven Tel, Zongwei Wu, Yulun Zhang, Barth ´el´emy Heyr- man, C ´edric Demonceaux, Radu Timofte, and Dominique Ginhac. Alignment-free hdr deghosting with semantics con- sistent transformer. InICCV, 2023. 2

    2023

  49. [49]

    Aim 2025 challenge on inverse tone mapping report: Methods and results

    Chao Wang, Francesco Banterle, Bin Ren, Radu Timofte, Xin Lu, Yufeng Peng, Chengjie Ge, Zhijing Sun, Ziang Zhou, Zihao Li, et al. Aim 2025 challenge on inverse tone mapping report: Methods and results. InICCV, pages 5571– 5584, 2025. 5, 6, 8

    2025

  50. [50]

    Multi- scale structural similarity for image quality assessment

    Zhou Wang, Eero P Simoncelli, and Alan C Bovik. Multi- scale structural similarity for image quality assessment. In The thrity-seventh asilomar conference on signals, systems & computers, 2003, pages 1398–1402. IEEE, 2003. 6

    2003

  51. [51]

    Bovik, H.R

    Zhou Wang, A.C. Bovik, H.R. Sheikh, and E.P. Simoncelli. Image quality assessment: from error visibility to structural similarity.IEEE TIP, 13(4):600–612, 2004. 6

    2004

  52. [52]

    Deep high dynamic range imaging with large foreground motions

    Shangzhe Wu, Jiarui Xu, Yu-Wing Tai, and Chi-Keung Tang. Deep high dynamic range imaging with large foreground motions. InECCV, pages 117–132, 2018. 1, 2

    2018

  53. [53]

    Smartbrush: Text and shape guided object inpainting with diffusion model

    Shaoan Xie, Zhifei Zhang, Zhe Lin, Tobias Hinz, and Kun Zhang. Smartbrush: Text and shape guided object inpainting with diffusion model. InCVPR, pages 22428–22437, 2023. 3

    2023

  54. [54]

    Attention- guided network for ghost-free high dynamic range imaging

    Qingsen Yan, Dong Gong, Qinfeng Shi, Anton van den Hen- gel, Chunhua Shen, Ian Reid, and Yanning Zhang. Attention- guided network for ghost-free high dynamic range imaging. InCVPR, pages 1751–1760, 2019. 1, 2

    2019

  55. [55]

    The unreasonable effectiveness of deep features as a perceptual metric

    Richard Zhang, Phillip Isola, Alexei A Efros, Eli Shechtman, and Oliver Wang. The unreasonable effectiveness of deep features as a perceptual metric. InCVPR, pages 586–595,

  56. [56]

    Sine: Single image editing with text-to-image diffusion models

    Zhixing Zhang, Ligong Han, Arnab Ghosh, Dimitris N Metaxas, and Jian Ren. Sine: Single image editing with text-to-image diffusion models. InCVPR, pages 6027–6037,

  57. [57]

    Unmodnet: Learning to unwrap a modulo image for high dynamic range imaging

    Chu Zhou, Hang Zhao, Jin Han, Chang Xu, Chao Xu, Tiejun Huang, and Boxin Shi. Unmodnet: Learning to unwrap a modulo image for high dynamic range imaging. InNeurIPS, pages 1559–1570, 2020. 2

    2020

  58. [58]

    Recdiffu- sion: Rectangling for image stitching with diffusion models

    Tianhao Zhou, Haipeng Li, Ziyi Wang, Ao Luo, Chen-Lin Zhang, Jiajun Li, Bing Zeng, and Shuaicheng Liu. Recdiffu- sion: Rectangling for image stitching with diffusion models. InCVPR, pages 2692–2701, 2024. 3

    2024