MESS: Multi-Exposure Sequence Synthesis for Generalizable Image Enhancement

Ruodai Cui; Shuaizheng Liu; Rongyuan Wu; Lei Zhang

cs.CVOct 17, 2026

MESS: Multi-Exposure Sequence Synthesis for Generalizable Image Enhancement

Authors: Ruodai Cui, Shuaizheng Liu, Rongyuan Wu, Lei Zhang

Organizations: Department of Computing, The Hong Kong Polytechnic University · Qualcomm Technologies, Inc. · OPPO Research Institute

Abstract

Despite the significant progress in image enhancement under adverse illuminations, such as low-light image enhancement, exposure correction, and backlit image enhancement, existing models show limited generalizability to different scenarios. A major bottleneck lies in the lack of large-scale and diverse training data, as these tasks typically rely on manually captured, precisely aligned image pairs that are expensive to scale up. Although synthetic data have been explored as an alternative, existing synthesis methods are limited in modeling complex adverse illumination conditions and camera imaging pipelines. To address this issue, we propose a novel framework to synthesize realistic training pairs for generalizable image enhancement in the wild. Specifically, we first derive multi-exposure sequences (MES) from high-bit RAW images by rendering them under different exposure settings through an emulated ISP pipeline. Using these RAW-derived MES as supervision, we train a diffusion-based RGB-to-MES generator to synthesize plausible exposure sequences from a single 8-bit RGB image, extending the synthesis pipeline from limited RAW collections to large-scale in-the-wild RGB data. For each source image, we randomly sample one synthesized exposure variant as the degraded input and use the original high-quality RGB image as the target. With the proposed MES Synthesis (MESS) approach, we construct a dataset of 500K training pairs and train a lightweight network, which demonstrates significantly better generalization performance than existing models across various image enhancement tasks. Code and data are available at https://github.com/PolyU-VCLab/MESS.

Figures & tables

Panel of Figure 1 — Figure 1: Visual comparisons of a recent state-of-the-art low-light image enhancement method, Rformer [ 8 ] , trained on the UHDLL dataset [ 35 ] (2nd column) and on our large-scale synthetic dataset (3rd column). Zoomed-in regions are highlighted with red boxes for better comparison.

Figure 2: Comparison of different data construction strategies for image illumination enhancement. RAW-based methods (top left) produce realistic exposure variations but lack scalability, while RGB-based methods (top right) are scalable but often less realistic. In contrast, our proposed MESS (bottom) transfers RAW-derived exposure priors to in-the-wild RGB images, enabling both realistic and scalable synthesis of training pairs. Notions are defined in § 3 .

Task	Dataset	#images	#scenes	Task	Dataset	#images	#scenes
LLIE	LOLv2Real [ 75 ]	789	263	EC	LCDP [ 62 ]	1,700	1,700
LLIE	LSRW [ 22 ]	5,650	4,157	EC	SICE [ 7 ]	4,800	589
LLIE	UHD-LL [ 35 ]	2,150	724	BIE	BAID [ 48 ]	3,000	1,782
EC	MSEC [ 1 ]	24,330	5,000	BIE	Backlit300 [ 42 ]	305	305

Table 1: Summary of existing datasets for image enhancement. “#images” denotes the number of training images, while “#scenes” denotes the number of unique scenes. For datasets LOLv2Real [ 75 ] , UHD-LL [ 35 ] , LSRW [ 22 ] , and BAID [ 48 ] , the exact number of scenes is not provided. We treat images with a CLIP similarity score greater than 0.9 as duplicated images from the same scene.

Figure 3: MESS Training Framework. (a) The training pipeline: Given a single RGB image and a target exposure value, MESS generates the corresponding exposure-adjusted image through three modules: frequency-based enhancement, diffusion-based synthesis, and exposure-conditioned rendering. (b) Structure of the frequency-based enhancement module. (c) Structure of the exposure-conditioned rendering module.

Training Data	Type	Metric	UHDLL	LSRW	SICE	MSEC	BAID	Average
UHDLL [ 35 ]	Real	PSNR $\uparrow$	23.694	19.357	16.152	18.728	20.461	19.678
LSRW [ 22 ]	Real	PSNR $\uparrow$	21.383	20.849	17.246	18.964	19.875	19.663
SICE [ 7 ]	Real	PSNR $\uparrow$	18.947	18.061	22.385	20.329	20.103	19.965
MSEC [ 1 ]	Raw-Syn	PSNR $\uparrow$	19.348	18.214	20.367	22.116	20.352	20.079
BAID [ 48 ]	Real	PSNR $\uparrow$	17.684	16.935	18.276	18.742	21.674	18.662
RCD [ 11 ]	Raw-Syn	PSNR $\uparrow$	19.702	18.242	17.478	19.924	20.159	19.101

Table 2: The results of full-reference scores using FourLLIE backbone [ 61 ] . The model trained on each data source are evaluated on all test sets without fine-tuning. Values in red and blue indicate the best and second-best results, respectively.

Training Data	Type	Metric	UHDLL	LSRW	SICE	MSEC	BAID	Average
UHDLL [ 35 ]	Real	ARNIQA $\uparrow$	0.681	0.650	0.694	0.693	0.703	0.684
LSRW [ 22 ]	Real	ARNIQA $\uparrow$	0.662	0.671	0.688	0.691	0.699	0.682
SICE [ 7 ]	Real	ARNIQA $\uparrow$	0.666	0.630	0.720	0.711	0.693	0.684
MSEC [ 1 ]	Raw-Syn	ARNIQA $\uparrow$	0.663	0.648	0.719	0.720	0.708	0.692
BAID [ 48 ]	Real	ARNIQA $\uparrow$	0.628	0.611	0.652	0.667	0.727	0.657
RCD [ 11 ]	Raw-Syn	ARNIQA $\uparrow$	0.663	0.640	0.693	0.697	0.699	0.678

Table 3: The results of no-reference scores using FourLLIE backbone [ 61 ] . The model trained on each data source are evaluated on all test sets without fine-tuning. Values in red and blue indicate the best and second-best results, respectively.

Figure 4: Qualitative comparison of the FourLLIE models trained on UHDLL, LSRW and our MESS generated datasets. The test image is from the LSRW dataset.

Figure 5: Qualitative comparison of the FourLLIE models trained on MSEC, UEC and MESS generated datasets. The test image is from the SICE dataset.

Training data	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	ARNIQA $\uparrow$	MANIQA $\uparrow$	MUSIQ $\uparrow$
LSRW	19.663	0.816	0.149	0.682	0.444	53.508
SICE	19.965	0.816	0.149	0.684	0.444	53.752
MESS-5K	20.220	0.822	0.143	0.686	0.451	54.050
MESS-10K	20.590	0.830	0.137	0.692	0.456	54.410
MESS-50K	21.199	0.839	0.129	0.701	0.463	54.988
MESS-500K	21.923	0.846	0.121	0.709	0.467	55.510

Table 4: Dataset-scale comparison using FourLLIE. Scores are macro-averaged over the five test sets. MESS-5K is approximately comparable in size to LSRW and SICE. Values in red and blue indicate the best and second-best results, respectively.

Training data	Unique scenes	Pairs	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$
RAW-only	39,642	39,642	20.764	0.829	0.139
MESS-RGB, matched size	39,642	39,642	21.083	0.837	0.131
MESS-RGB, full set	500,000	500,000	21.923	0.846	0.121

Table 5: RAW-only and MESS-RGB comparison. Results are macro-averaged over UHDLL, LSRW, SICE, MSEC, and BAID. Values in red and blue indicate the best and second-best results, respectively.

Panel of Figure 6 — Figure 6: Visualization of MESS synthesis. The source RGB image is followed by synthesized exposure variants from MESS at $-3$ , $-1$ , $+1$ , and $+3$ EV, respectively.

Appendix figures & tables18 assets

Supplementary material from the paper’s appendix.

Appendix

Training Data	Type	Metric	UHDLL	LSRW	SICE	MSEC	BAID
UHDLL [ 35 ]	Real	PSNR $\uparrow$	22.509	19.024	15.687	18.745	20.025
LSRW [ 22 ]	Real	PSNR $\uparrow$	20.175	20.496	16.745	18.980	19.440
SICE [ 7 ]	Real	PSNR $\uparrow$	17.770	17.711	21.899	20.300	19.645
MSEC [ 1 ]	Raw-Syn	PSNR $\uparrow$	18.162	17.890	19.910	22.809	19.910
BAID [ 48 ]	Real	PSNR $\uparrow$	16.508	16.591	17.782	18.750	21.471
RCD [ 11 ]	Raw-Syn	PSNR $\uparrow$	18.701	17.855	18.641	20.768	20.243

Appendix

Table 6: Full-reference results using RetinexFormer [ 8 ] backbone. The model trained on each data source is evaluated on all test sets without fine-tuning. Values in red and blue indicate the best and second-best results, respectively.

Training Data	Type	Metric	UHDLL	LSRW	SICE	MSEC	BAID
UHDLL [ 35 ]	Real	ARNIQA $\uparrow$	0.667	0.657	0.693	0.694	0.680
LSRW [ 22 ]	Real	ARNIQA $\uparrow$	0.648	0.685	0.685	0.692	0.678
SICE [ 7 ]	Real	ARNIQA $\uparrow$	0.651	0.643	0.717	0.711	0.671
MSEC [ 1 ]	Raw-Syn	ARNIQA $\uparrow$	0.647	0.658	0.707	0.713	0.684
BAID [ 48 ]	Real	ARNIQA $\uparrow$	0.613	0.626	0.650	0.666	0.705
RCD [ 11 ]	Raw-Syn	ARNIQA $\uparrow$	0.641	0.651	0.691	0.708	0.681

Appendix

Table 7: No-reference results using RetinexFormer [ 8 ] backbone. The model trained on each data source is evaluated on all test sets without fine-tuning. Values in red and blue indicate the best and second-best results, respectively.

Training Data	Type	Metric	UHDLL	LSRW	SICE	MSEC	BAID
UHDLL [ 35 ]	Real	PSNR $\uparrow$	20.622	17.579	14.033	18.743	18.473
LSRW [ 22 ]	Real	PSNR $\uparrow$	18.295	19.061	15.103	18.971	17.878
SICE [ 7 ]	Real	PSNR $\uparrow$	15.878	16.274	20.073	20.302	18.084
MSEC [ 1 ]	Raw-Syn	PSNR $\uparrow$	16.274	16.448	18.265	22.322	18.345
BAID [ 48 ]	Real	PSNR $\uparrow$	14.611	15.146	16.149	18.748	19.918
RCD [ 11 ]	Raw-Syn	PSNR $\uparrow$	16.818	16.417	16.995	20.769	18.685

Appendix

Table 8: Full-reference results using Multinex [ 4 ] backbone. The model trained on each data source is evaluated on all test sets without fine-tuning. Values in red and blue indicate the best and second-best results, respectively.

Training Data	Type	Metric	UHDLL	LSRW	SICE	MSEC	BAID
UHDLL [ 35 ]	Real	ARNIQA $\uparrow$	0.614	0.622	0.672	0.692	0.646
LSRW [ 22 ]	Real	ARNIQA $\uparrow$	0.595	0.645	0.666	0.689	0.639
SICE [ 7 ]	Real	ARNIQA $\uparrow$	0.598	0.605	0.699	0.710	0.637
MSEC [ 1 ]	Raw-Syn	ARNIQA $\uparrow$	0.612	0.618	0.690	0.702	0.647
BAID [ 48 ]	Real	ARNIQA $\uparrow$	0.559	0.585	0.631	0.675	0.669
RCD [ 11 ]	Raw-Syn	ARNIQA $\uparrow$	0.602	0.615	0.672	0.703	0.653

Appendix

Table 9: No-reference results using Multinex [ 4 ] backbone. The model trained on each data source is evaluated on all test sets without fine-tuning. Values in red and blue indicate the best and second-best results, respectively.

Training Data	Type	Metric	UHDLL	LSRW	SICE	MSEC	BAID
UHDLL [ 35 ]	Real	PSNR $\uparrow$	19.254	16.508	12.332	17.656	16.346
LSRW [ 22 ]	Real	PSNR $\uparrow$	16.957	17.977	13.427	18.935	15.748
SICE [ 7 ]	Real	PSNR $\uparrow$	14.518	15.173	17.934	20.345	16.013
MSEC [ 1 ]	Raw-Syn	PSNR $\uparrow$	14.927	15.361	16.544	23.200	16.222
BAID [ 48 ]	Real	PSNR $\uparrow$	13.270	14.047	14.439	18.762	17.823
RCD [ 11 ]	Raw-Syn	PSNR $\uparrow$	15.469	15.322	15.283	20.558	16.564

Appendix

Table 10: Full-reference results using ExSLOT [ 29 ] backbone. The model trained on each data source is evaluated on all test sets without fine-tuning. Values in red and blue indicate the best and second-best results, respectively.

Training Data	Type	Metric	UHDLL	LSRW	SICE	MSEC	BAID
UHDLL [ 35 ]	Real	ARNIQA $\uparrow$	0.661	0.650	0.693	0.692	0.680
LSRW [ 22 ]	Real	ARNIQA $\uparrow$	0.640	0.673	0.690	0.690	0.675
SICE [ 7 ]	Real	ARNIQA $\uparrow$	0.646	0.633	0.727	0.709	0.669
MSEC [ 1 ]	Raw-Syn	ARNIQA $\uparrow$	0.641	0.652	0.720	0.728	0.687
BAID [ 48 ]	Real	ARNIQA $\uparrow$	0.610	0.613	0.654	0.669	0.705
RCD [ 11 ]	Raw-Syn	ARNIQA $\uparrow$	0.639	0.649	0.707	0.707	0.682

Appendix

Table 11: No-reference results using ExSLOT [ 29 ] backbone. The model trained on each data source is evaluated on all test sets without fine-tuning. Values in red and blue indicate the best and second-best results, respectively.

Training Data	Type	Metric	UHDLL	LSRW	SICE	MSEC	BAID
UHDLL [ 35 ]	Real	PSNR $\uparrow$	19.145	16.386	11.687	18.706	15.269
LSRW [ 22 ]	Real	PSNR $\uparrow$	16.860	17.850	12.807	18.961	14.696
SICE [ 7 ]	Real	PSNR $\uparrow$	14.421	15.039	17.321	20.334	14.905
MSEC [ 1 ]	Raw-Syn	PSNR $\uparrow$	14.824	15.197	15.910	22.728	15.169
BAID [ 48 ]	Real	PSNR $\uparrow$	12.278	13.928	13.841	18.745	16.763
RCD [ 11 ]	Raw-Syn	PSNR $\uparrow$	15.177	15.184	14.654	20.761	15.508

Appendix

Table 12: Full-reference results using CSEC [ 38 ] backbone. The model trained on each data source is evaluated on all test sets without fine-tuning. Values in red and blue indicate the best and second-best results, respectively.

Training Data	Type	Metric	UHDLL	LSRW	SICE	MSEC	BAID
UHDLL [ 35 ]	Real	ARNIQA $\uparrow$	0.682	0.659	0.708	0.692	0.700
LSRW [ 22 ]	Real	ARNIQA $\uparrow$	0.661	0.682	0.701	0.692	0.695
SICE [ 7 ]	Real	ARNIQA $\uparrow$	0.669	0.643	0.734	0.709	0.690
MSEC [ 1 ]	Raw-Syn	ARNIQA $\uparrow$	0.664	0.657	0.725	0.722	0.701
BAID [ 48 ]	Real	ARNIQA $\uparrow$	0.629	0.623	0.668	0.667	0.734
RCD [ 11 ]	Raw-Syn	ARNIQA $\uparrow$	0.664	0.655	0.719	0.706	0.705

Appendix

Table 13: No-reference results using CSEC [ 38 ] backbone. The model trained on each data source is evaluated on all test sets without fine-tuning. Values in red and blue indicate the best and second-best results, respectively.

Training Data	Type	Metric	UHDLL	LSRW	SICE	MSEC	BAID
UHDLL [ 35 ]	Real	PSNR $\uparrow$	19.276	16.488	12.454	18.718	16.573
LSRW [ 22 ]	Real	PSNR $\uparrow$	16.983	18.003	13.557	18.981	15.976
SICE [ 7 ]	Real	PSNR $\uparrow$	14.521	15.214	18.673	20.311	16.182
MSEC [ 1 ]	Raw-Syn	PSNR $\uparrow$	14.922	15.362	16.650	22.001	16.473
BAID [ 48 ]	Real	PSNR $\uparrow$	13.249	14.094	14.574	18.772	18.035
RCD [ 11 ]	Raw-Syn	PSNR $\uparrow$	15.459	15.332	15.398	20.761	16.803

Appendix

Table 14: Full-reference results using RAVE [ 17 ] backbone. The model trained on each data source is evaluated on all test sets without fine-tuning. Values in red and blue indicate the best and second-best results, respectively.

Training Data	Type	Metric	UHDLL	LSRW	SICE	MSEC	BAID
UHDLL [ 35 ]	Real	ARNIQA $\uparrow$	0.657	0.650	0.692	0.693	0.679
LSRW [ 22 ]	Real	ARNIQA $\uparrow$	0.637	0.672	0.687	0.692	0.672
SICE [ 7 ]	Real	ARNIQA $\uparrow$	0.642	0.633	0.718	0.699	0.667
MSEC [ 1 ]	Raw-Syn	ARNIQA $\uparrow$	0.638	0.645	0.708	0.697	0.680
BAID [ 48 ]	Real	ARNIQA $\uparrow$	0.603	0.613	0.649	0.668	0.701
RCD [ 11 ]	Raw-Syn	ARNIQA $\uparrow$	0.632	0.643	0.691	0.702	0.689

Appendix

Table 15: No-reference results using RAVE [ 17 ] backbone. The model trained on each data source is evaluated on all test sets without fine-tuning. Values in red and blue indicate the best and second-best results, respectively.

Panel of Figure 7 — Figure 7: Qualitative comparison of the FourLLIE models trained on UHDLL, LSRW, and our MESS-generated datasets. The test images are from the BAID dataset.

Panel of Figure 8 — Figure 8: Qualitative comparison of the FourLLIE models trained on UHDLL, LSRW, and our MESS-generated datasets. The test images are from the LSRW dataset.

Panel of Figure 9 — Figure 9: Qualitative comparison of the FourLLIE models trained on UHDLL, LSRW, and our MESS-generated datasets. The test images are from the SICE dataset.

Panel of Figure 10 — Figure 10: Qualitative comparison of the FourLLIE models trained on MSEC, UEC, and MESS-generated datasets. The test images are from the BAID dataset.

Panel of Figure 11 — Figure 11: Qualitative comparison of the FourLLIE models trained on MSEC, UEC, and MESS-generated datasets. The test images are from the SICE dataset.

Panel of Figure 12 — Figure 12: Qualitative comparison of the FourLLIE models trained on MSEC, UEC, and MESS-generated datasets. The test images are from the LSRW dataset.

Variant	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	ARNIQA $\uparrow$	MUSIQ $\uparrow$	MANIQA $\uparrow$
w/o representation consistency loss	21.354	0.835	0.132	0.698	54.821	0.456
w/o frequency-based enhancement	20.936	0.828	0.140	0.692	54.307	0.449
w/o diffusion-based detail enhancement	21.112	0.832	0.136	0.695	54.563	0.453
w/o exposure-conditioned rendering	20.781	0.821	0.145	0.688	54.018	0.445
FFT instead of DWT	21.447	0.837	0.130	0.700	54.963	0.458
Full MESS	21.923	0.846	0.121	0.709	55.510	0.467

Appendix

Table 16: Ablation study of the proposed MESS framework. All variants are used to generate training data for the same FourLLIE backbone, and the trained models are evaluated on five test sets without fine-tuning. The reported results are averaged over UHDLL, LSRW, SICE, MSEC, and BAID.

Renderer Design	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	ARNIQA $\uparrow$	MUSIQ $\uparrow$	MANIQA $\uparrow$
No exposure condition	20.781	0.821	0.145	0.688	54.018	0.445
Concatenate EV with RGB features	21.082	0.831	0.137	0.695	54.576	0.452
Global FiLM on RGB features	21.324	0.836	0.132	0.701	54.905	0.459
DWT-FiLM w/o content descriptor	21.541	0.840	0.128	0.704	55.102	0.462
DWT-FiLM with content descriptor	21.923	0.846	0.121	0.709	55.510	0.467

Appendix

Table 17: Ablation study on the design of the exposure-conditioned rendering module. All variants use the same frequency-based enhancement module and diffusion-based detail enhancement module. Only the way of injecting the target exposure condition is changed.

Image Enhancement

Image enhancement restores visibility, color, detail, and dynamic range in degraded images for human viewing and downstream vision tasks.

See all 154 papers

RawGen: Learning Camera Raw Image Generation
Mar 31, 2026Dongyoung Kim, Junyong Lee, Abhijith Punnappurath +4Computational Imaging Image Generation
AI Center - Toronto, Samsung Electronics · Yonsei University
UBLLIE: Unified Backlight and Low-Light Image Enhancement
Aug 5, 2026Yasmin Yasin, Muhammad Usman, Ibrahim Radwan +1Image Enhancement Low-Light Image Enhancement
King Fahd University of Petroleum and Minerals, Dhahran, Saudi Arabia · Ontario Tech University, Oshawa, Canada · University of Western Australia, Perth, Australia +1
GGT-100K: Generative Ground Truth for Generalizable Real-World Image Restoration
May 29, 2026Xiangtao Kong, Jixin Zhao, Lingchen Sun +2Synthetic Data Augmentation Image Restoration
The Hong Kong Polytechnic University · OPPO Research Institute

Mar 31, 2026J/K move · Enter open · S save

RawGen: Learning Camera Raw Image Generation

Dongyoung Kim, Junyong Lee, Abhijith Punnappurath, Mahmoud Afifi+3

AI Center - Toronto, Samsung Electronics · Yonsei University

Computational Imaging Image Generation Diffusion-Based Image Generation Inverse Problems T2I Diffusion Models

Open paper AskPDF arXiv

Cameras capture scene-referred linear raw images, which are processed by onboard image signal processors (ISPs) into display-referred 8-bit sRGB outputs. Although raw data is more faithful for low-level vision tasks, collecting large-scale raw datasets remains a major bottleneck, as existing datasets are limited and tied to specific camera hardware. Generative models offer a promising way to address this scarcity -- however, existing diffusion frameworks are designed to synthesize photo-finished sRGB images rather than physically consistent linear representations. This paper presents RawGen, to our knowledge the first diffusion-based framework enabling text-to-raw generation for arbitrary target cameras, alongside sRGB-to-raw inversion. RawGen leverages the generative priors of large-scale sRGB diffusion models to synthesize physically meaningful linear outputs, such as CIE XYZ or camera-specific raw representations, via specialized processing in latent and pixel spaces. To handle unknown and diverse ISP pipelines and photo-finishing effects in diffusion-model training data, we build a many-to-one inverse-ISP dataset where multiple sRGB renditions of the same scene generated using diverse ISP parameters are anchored to a common scene-referred target. Fine-tuning a conditional denoiser and specialized decoder on this dataset allows RawGen to obtain camera-centric linear reconstructions that effectively invert the rendering pipeline. We demonstrate RawGen's superior performance over traditional inverse-ISP methods that assume a fixed ISP. Furthermore, we show that augmenting training pipelines with RawGen's scalable, text-driven synthetic data can benefit downstream low-level vision tasks.

Abstract

Image Enhancement

Explore similar work

RawGen: Learning Camera Raw Image Generation

UBLLIE: Unified Backlight and Low-Light Image Enhancement

GGT-100K: Generative Ground Truth for Generalizable Real-World Image Restoration