Journal of Zhejiang University

ENGINEERING Information Technology & Electronic Engineering 2026 Vol.27 No.6 P.1-14

http://doi.org/10.1631/ENG.ITEE.2025.0116

Unsupervised single-image high dynamic range rendering via multi-exposure priors

Author(s): Han WANG, Bolun ZHENG, Quan CHEN, Qianyu ZHANG, Tao ZHANG, Jiyong ZHANG, Xiang TIAN
Affiliation(s): 1. School of Automation, Hangzhou Dianzi University, Hangzhou 310018, China more
Corresponding email(s): chenquan@alu.hdu.edu.cn
Key Words: High dynamic range (HDR), HDR reconstruction, Single-image HDR, Unsupervised learning, Multi-exposure prior

Share this article to： More <<< Previous Article \|Next Article >>>

Han WANG, Bolun ZHENG, Quan CHEN, Qianyu ZHANG, Tao ZHANG, Jiyong ZHANG, Xiang TIAN. Unsupervised single-image high dynamic range rendering via multi-exposure priors[J]. Journal of Zhejiang University Science C, 2026, 27(6): 1-14.

@article{title="Unsupervised single-image high dynamic range rendering via multi-exposure priors",
author="Han WANG, Bolun ZHENG, Quan CHEN, Qianyu ZHANG, Tao ZHANG, Jiyong ZHANG, Xiang TIAN",
journal="Journal of Zhejiang University Science C",
volume="27",
number="6",
pages="1-14",
year="2026",
publisher="Zhejiang University Press & Springer",
doi="10.1631/ENG.ITEE.2025.0116"
}

%0 Journal Article
%T Unsupervised single-image high dynamic range rendering via multi-exposure priors
%A Han WANG
%A Bolun ZHENG
%A Quan CHEN
%A Qianyu ZHANG
%A Tao ZHANG
%A Jiyong ZHANG
%A Xiang TIAN
%J Frontiers of Information Technology & Electronic Engineering
%V 27
%N 6
%P 1-14
%@ 1869-1951
%D 2026
%I Zhejiang University Press & Springer
%DOI 10.1631/ENG.ITEE.2025.0116

TY - JOUR
T1 - Unsupervised single-image high dynamic range rendering via multi-exposure priors
A1 - Han WANG
A1 - Bolun ZHENG
A1 - Quan CHEN
A1 - Qianyu ZHANG
A1 - Tao ZHANG
A1 - Jiyong ZHANG
A1 - Xiang TIAN
J0 - Frontiers of Information Technology & Electronic Engineering
VL - 27
IS - 6
SP - 1
EP - 14
%@ 1869-1951
Y1 - 2026
PB - Zhejiang University Press & Springer
ER -
DOI - 10.1631/ENG.ITEE.2025.0116

Abstract
Chinese Summary
Academic Network
Reviewer Comment

Abstract: Reconstructing high dynamic range (HDR) images from a single low dynamic range (LDR) input requires recovering missing information in highlight-clipped and shadow-distorted regions. Existing methods generally rely on sufficient ground-truth HDR images as supervision signals or multi-exposure LDR sequences to improve quality, limiting their flexibility. To address this, we propose USME-HDR, a framework for single-image HDR reconstruction based on multi-exposure priors, where the HDR reconstruction stage is learned without ground-truth HDR supervision. Specifically, an exposure-adjustment network (EAN) is trained in a supervised manner to map a single LDR image to over/under-exposure pairs. Inspired by the Retinex theory, we further decompose the input into a light map and a light feature, which are fed into the EAN as auxiliary inputs for luminance-aware exposure generation. An exposure time ratio guidance mechanism is further introduced to improve luminance fidelity. Finally, the HDR image is synthesized by fusing the original LDR image with generated multi-exposure images, refined through self-supervised optimization. Experiments demonstrate that during the test phase, USME-HDR reconstructs visually compelling HDR images from only a single LDR input, without requiring real low- or high-exposure images.

基于多曝光先验的无监督单图像高动态范围渲染

王涵¹，郑博仑¹，陈泉²，张仟瑜¹，张涛¹，张继勇¹，田翔³
¹杭州电子科技大学自动化学院，中国杭州市，310018
²嘉兴大学人工智能学院，中国嘉兴市，314001
²浙江大学先进数字技术与仪器研究所，中国杭州市，310027
摘要：从单张低动态范围图像重建高动态范围图像，需要恢复高光饱和区域和阴影失真区域中缺失的信息。现有方法通常依赖充足的真实高动态范围图像作为监督信号，或利用多曝光低动态范围序列以提升重建质量，在一定程度上限制了方法的灵活性。为此，提出一种基于多曝光先验的单图像高动态范围重建框架USME-HDR，其中高动态范围重建阶段的学习无需真实高动态范围图像的监督。具体而言，首先以监督方式训练一个曝光调整网络，将单张低动态范围图像映射为过曝/欠曝图像对。受Retinex理论启发，进一步将输入图像分解为光照图和光照特征，并将其作为辅助输入引入曝光调整网络，以实现基于亮度感知的曝光生成。同时，引入曝光时间比引导机制以提升亮度保真度。最后，通过融合原始低动态范围图像与生成的多曝光图像，并结合自监督优化过程，生成最终的高动态范围图像。实验结果表明，在测试阶段，USME-HDR仅需单张低动态范围图像输入即可重建出具有良好视觉效果的高动态范围图像，无需真实的低曝光或高曝光图像。

关键词：高动态范围；高动态范围重建；单图像高动态范围；无监督学习；多曝光先验

Darkslateblue:Affiliate; Royal Blue:Author; Turquoise:Article

Reference

[1]Bai JS, Yin YH, He QY, et al., 2024. RetinexMamba: Retinex-based Mamba for low-light image enhancement. Proc 31^st Int Conf on Neural Information Processing, p.427-442.

[2]Cai YH, Bian H, Lin J, et al., 2023. Retinexformer: one-stage Retinex-based Transformer for low-light image enhancement. Proc IEEE/CVF Int Conf on Computer Vision, p.12470-12479.

[3]Chen RF, Zheng BL, Zhang H, et al., 2023. Improving dynamic HDR imaging with fusion transformer. Proc AAAI Conf on Artificial Intelligence, p.340-349.

[4]Chen SK, Yen HL, Liu YL, et al., 2023. Learning continuous exposure value representations for single-image HDR reconstruction. Proc IEEE/CVF Int Conf on Computer Vision, p.12944-12954.

[5]Chen XY, Liu YH, Zhang ZW, et al., 2021. HDRUNet: single image HDR reconstruction with denoising and dequantization. Proc IEEE/CVF Conf on Computer Vision and Pattern Recognition Workshops, p.354-363.

[6]Chen ZX, Wang YJ, Cai X, et al., 2025. UltraFusion: ultra high dynamic imaging using exposure fusion. Proc IEEE/CVF Conf on Computer Vision and Pattern Recognition, p.16111-16121.

[7]Debevec PE, Malik J, 1997. Recovering high dynamic range radiance maps from photographs. Proc 24^th Annual Conf on Computer Graphics and Interactive Techniques, p.369-378.

[8]Dille S, Careaga C, Aksoy Y, 2024. Intrinsic single-image HDR reconstruction. Proc 18^th European Conf on Computer Vision, p.161-177.

[9]Endo Y, Kanamori Y, Mitani J, 2017. Deep reverse tone mapping. ACM Trans Graph, 36(6):177.

[10]Hu T, Yan QS, Qi YK, et al., 2024. Generating content for HDR deghosting from frequency view. Proc IEEE/CVF Conf on Computer Vision and Pattern Recognition, p.25732-25741.

[11]Huang X, Zhang Q, Feng Y, et al., 2022. HDR-NeRF: high dynamic range neural radiance fields. Proc IEEE/CVF Conf on Computer Vision and Pattern Recognition, p.18377-18387.

[12]Kalantari NK, Ramamoorthi R, 2017. Deep high dynamic range imaging of dynamic scenes. ACM Trans Graph, 36(4):144.

[13]Ke JJ, Wang QF, Wang YL, et al., 2021. MUSIQ: multi-scale image quality Transformer. Proc IEEE/CVF Int Conf on Computer Vision, p.5128-5137.

[14]Khan Z, Khanna M, Raman S, 2019. FHDR: HDR image reconstruction from a single LDR image using feedback network. Proc IEEE Global Conf on Signal and Information Processing, p.1-5.

[15]Kong LT, Li B, Xiong YK, et al., 2024. SAFNet: selective alignment fusion network for efficient HDR imaging. Proc 18^th European Conf on Computer Vision, p.256-273.

[16]Land EH, McCann JJ, 1971. Lightness and retinex theory. J Opt Soc Am, 61(1):1-11.

[17]Le PH, Le Q, Nguyen R, et al., 2023. Single-image HDR reconstruction by multi-exposure generation. Proc IEEE/CVF Winter Conf on Applications of Computer Vision, p.4052-4061.

[18]Lee S, An GH, Kang SJ, 2018. Deep recursive HDRI: inverse tone mapping using generative adversarial networks. Proc 15^th European Conf on Computer Vision, p.613-628.

[19]Li ZW, Zhang F, Cao M, et al., 2024. Real-time exposure correction via collaborative transformations and adaptive sampling. Proc IEEE/CVF Conf on Computer Vision and Pattern Recognition, p.2984-2994.

[20]Liu SZ, Zhang XD, Sun LC, et al., 2023. Joint HDR denoising and fusion: a real-world mobile HDR image dataset. Proc IEEE/CVF Conf on Computer Vision and Pattern Recognition, p.13966-13975.

[21]Liu YL, Lai WS, Chen YS, et al., 2020. Single-image HDR reconstruction by learning to reverse the camera pipeline. Proc IEEE/CVF Conf on Computer Vision and Pattern Recognition, p.1648-1657.

[22]Liu Z, Wang YL, Zeng B, et al., 2022. Ghost-free high dynamic range imaging with context-aware Transformer. Proc 17^th European Conf on Computer Vision, p.344-360.

[23]Mantiuk R, Kim KJ, Rempel AG, et al., 2011. HDR-VDP-2: a calibrated visual metric for visibility and quality predictions in all luminance conditions. ACM Trans Graph, 30(4):40.

[24]Mittal A, Moorthy AK, Bovik AC, 2012. No-reference image quality assessment in the spatial domain. IEEE Trans Image Process, 21(12):4695-4708.

[25]Mittal A, Soundararajan R, Bovik AC, 2013. Making a “completely blind” image quality analyzer. IEEE Signal Process Lett, 20(3):209-212.

[26]Nazarczuk M, Catley-Chandar S, Leonardis A, et al., 2024. Self-supervised HDR imaging from motion and exposure cues. Proc 18^th European Conf on Computer Vision, p.363-380.

[27]Prabhakar KR, Senthil G, Agrawal S, et al., 2021. Labeled from unlabeled: exploiting unlabeled data for few-shot deep HDR deghosting. Proc IEEE/CVF Conf on Computer Vision and Pattern Recognition, p.4873-4883.

[28]Santos MS, Ren TI, Kalantari NK, 2020. Single image HDR reconstruction using a CNN with masked features and perceptual loss. ACM Trans Graph, 39(4):1-10.

[29]Shu Y, Shen L, Hu X, et al., 2024. Towards real-world HDR video reconstruction: a large-scale benchmark dataset and a two-stage alignment network. Proc IEEE/CVF Conf on Computer Vision and Pattern Recognition, p.2879-2888.

[30]Simonyan K, Zisserman A, 2015. Very deep convolutional networks for large-scale image recognition. Proc 3^rd Int Conf on Learning Representations, p.1-14.

[31]Song JW, Park YI, Kong K, et al., 2022. Selective TransHDR: transformer-based selective HDR imaging using ghost region mask. Proc 17^th European Conf on Computer Vision, p.288-304.

[32]Tel S, Wu ZW, Zhang YL, et al., 2023. Alignment-free HDR deghosting with semantics consistent transformer. Proc IEEE/CVF Int Conf on Computer Vision, p.12790-12799.

[33]Wang H, Ye M, Zhu X, et al., 2022. KUNet: imaging knowledge-inspired single HDR image reconstruction. Proc 31^st Int Joint Conf on Artificial Intelligence, p.1408-1414.

[34]Wang JY, Chan KCK, Loy CC, 2023. Exploring CLIP for assessing the look and feel of images. Proc AAAI Conf on Artificial Intelligence, p.2555-2563.

[35]Wei C, Wang WJ, Yang WH, et al., 2018. Deep retinex decomposition for low-light enhancement. Proc British Machine Vision Conf, Article 155.

[36]Wu GY, Fu HM, Liu JY, et al., 2024. Hybrid-supervised dual-search: leveraging automatic learning for loss-free multi-exposure image fusion. Proc AAAI Conf on Artificial Intelligence, p.5985-5993.

[37]Xu GW, Wang YJ, Gu JW, et al., 2024. HDRFlow: real-time HDR video reconstruction with large motions. Proc IEEE/CVF Conf on Computer Vision and Pattern Recognition, p.24851-24860.

[38]Yan QS, Gong D, Shi QF, et al., 2019. Attention-guided network for ghost-free high dynamic range imaging. Proc IEEE/CVF Conf on Computer Vision and Pattern Recognition, p.1751-1760.

[39]Yan QS, Zhang S, Chen WY, et al., 2023a. SMAE: few-shot learning for HDR deghosting with saturation-aware masked autoencoders. Proc IEEE/CVF Conf on Computer Vision and Pattern Recognition, p.5775-5784.

[40]Yan QS, Chen WY, Zhang S, et al., 2023b. A unified HDR imaging method with pixel and patch level. Proc IEEE/CVF Conf on Computer Vision and Pattern Recognition, p.22211-22220.

[41]Yan QS, Yang KZ, Hu T, et al., 2025a. From dynamic to static: stepwisely generate HDR image for ghost removal. IEEE Trans Circ Syst Video Technol, 35(2):1409-1421.

[42]Yan QS, Feng YX, Zhang C, et al., 2025b. HVI: a new color space for low-light image enhancement. Proc IEEE/CVF Conf on Computer Vision and Pattern Recognition, p.5678-5687.

[43]Yang SY, Gu Z, Hao WY, et al., 2025. Few-shot exemplar-driven inpainting with parameter-efficient diffusion fine-tuning. Front Inform Technol Electron Eng, 26(8):1428-1440.

[44]Yu FH, Gu JJ, Li ZY, et al., 2024. Scaling up to excellence: practicing model scaling for photo-realistic image restoration in the wild. Proc IEEE/CVF Conf on Computer Vision and Pattern Recognition, p.25669-25680.

[45]Zhang ZL, Wang HY, Liu S, et al., 2024. Self-supervised high dynamic range imaging with multi-exposure images in dynamic scenes. Proc 12^th Int Conf on Learning Representations, p.25867-25884.

[46]Zheng BL, Chen Q, Yuan SX, et al., 2022a. Constrained predictive filters for single image bokeh rendering. IEEE Trans Comput Imaging, 8:346-357.

[47]Zheng BL, Pan XK, Zhang H, et al., 2022b. DomainPlus: cross transform domain learning towards high dynamic range imaging. Proc 30^th ACM Int Conf on Multimedia, p.1954-1963.

[48]Zhu YM, Wang L, Yuan JY, et al., 2025. A ground-based dataset and diffusion model for on-orbit low-light image enhancement. Front Inform Technol Electron Eng, 26(7):1083-1098.

[49]Zou YH, Yan CG, Fu Y, 2023. RawHDR: high dynamic range image reconstruction from a single raw image. Proc IEEE/CVF Int Conf on Computer Vision, p.12300-12310.

Open peer comments: Debate/Discuss/Question/Opinion

<1>