Deep Light Pollution Removal in Night Cityscape Photographs

Hao Wang; Xiaolin Wu; Xi Zhang; Baoqing Sun

cs.CVApr 10, 2026

Deep Light Pollution Removal in Night Cityscape Photographs

Authors: Hao Wang, Xiaolin Wu, Xi Zhang, Baoqing Sun

Organizations: Shandong University, Qingdao, China · Southwest Jiaotong University, Chengdu, China · Tongji University, Shanghai, China

Abstract

Nighttime photography is severely degraded by light pollution induced by pervasive artificial lighting in urban environments. After long-range scattering and spatial diffusion, unwanted artificial light overwhelms natural night luminance, generates skyglow that washes out the view of stars and celestial objects and produces halos and glow artifacts around light sources. Unlike nighttime dehazing, which aims to improve detail legibility through thick air, the objective of light pollution removal is to restore the pristine night appearance by neutralizing the radiative footprint of ground lighting. In this paper we introduce a physically-based degradation model that adds to the previous ones for nighttime dehazing two critical aspects; (i) anisotropic spread of directional light sources, and (ii) skyglow caused by invisible surface lights behind skylines. In addition, we construct a training strategy that leverages large generative model and synthetic-real coupling to compensate for the scarcity of paired real data and enhance generalization. Extensive experiments demonstrate that the proposed formulation and learning framework substantially reduce light pollution artifacts and better recover authentic night imagery than prior nighttime restoration methods.

Figures & tables

Figure 1. Examples of light-polluted images and our restoration results, showing effective light pollution removal with a natural nighttime appearance. Examples of light pollution and restoration photographs.

Figure 2. Overview of the synthetic light-pollution generation pipeline used for dataset construction. From a clean image $J$ , we successively add skyline-induced sky glow $P_{\text{sky}}$ , component-wise ALSF-based directional scattering $P_{\text{ALSF}}$ , and APSF-based isotropic scattering $P_{\text{APSF}}$ to obtain the polluted image $I$ .

Figure 3. Examples of our Dataset. (a) Variations of one clean image with simulated light pollution under randomized APSF and ALSF parameters. (b) Additional synthetic polluted samples. (c) Real severely light-polluted nighttime images in the test set.

Figure 4. Illustration of skyline-induced sky glow, where hidden ground lights behind the skyline create a visible glow region in the sky.

Figure 5. Overview of our pipeline. The polluted image $I$ and clean target $J$ are encoded into latents; the noisy clean latent is concatenated with the polluted latent and fed into a Q-LoRA fine-tuned Flux.1-Kontext model to predict the clean latent.

Method	Task	PSNR	SSIM
Jin’s (ACM MM 23) ( Jin et al., 2023 )	Dehaze	20.7015	0.7757
Cong’s (CVPR 24) ( Cong et al., 2024 )	Dehaze	18.1638	0.5701
Lin’s (AAAI 25) ( Lin et al., 2025 )	Dehaze	16.6349	0.5942
Jin’s (ECCV 22) ( Jin et al., 2022 )	LES	19.3228	0.8020
Dille’s (ECCV 24) ( Dille et al., 2024 )	HDR	20.1549	0.7783
Dai’s (TPMAI 24) ( Dai et al., 2024 )	Deflare	19.3228	0.8713

Table 1. Objective metrics comparisons of ours and other methods. The best two scores are shown in Bold and underline . "LES" means "Light Effects Suppression" and "LPR" means "Light Pollution Removal".

Figure 6. Qualitative comparison of different methods. Specific tasks are marked in small grey text , "LES" means "Light Effects Suppression" and "LPR" means "Light Pollution Removal".

Figure 7. Backbone ablation showing better overexposed detail recovery with the generative backbone than with UNet.

Exp.	$\mathbf{P}_{\text{ALSF}}$	$\mathbf{P}_{\text{sky}}$	PSNR	SSIM
(a)	✗	✗	22.0269	0.8177
(b)	✓	✗	23.1589	0.8116
(c)	✗	✓	23.6253	0.8341
(d)	✓	✓	26.1728	0.8517

Table 2. Ablation studies of ALSF kernel and Sky Glow. The best two scores are shown in Bold and underline .

No.	Method	Average Rank
	Jin’s (ACM MM 23) ( Jin et al., 2023 )	4.67
	Cong’s (CVPR 24) ( Cong et al., 2024 )	6.38
	Lin’s (AAAI 25) ( Lin et al., 2025 )	5.11
	Jin’s (ECCV 22) ( Jin et al., 2022 )	5.18
	Dille’s (ECCV 24) ( Dille et al., 2024 )	3.92
	Dai’s (TPMAI 24) ( Dai et al., 2024 )	3.96

Table 3. Average Rankings of Methods. The best average ranks are shown in Bold .

Figure 8. Comparison with instruction-based image editing models, showing that task-specific fine-tuning is needed for severe light pollution removal.

Figure 9. Top-1 ranking rates for different methods.

Figure 10. Visualization of the ALSF construction process.

Appendix figures & tables5 assets

Supplementary material from the paper’s appendix.

Appendix

Figure 11. 3D bar chart illustrating the distribution of ranks assigned to each method in the user study.

Figure 12. Qualitative comparison of different methods. Specific tasks are marked in small grey text , "LES" means "Light Effects Suppression" and "LPR" means "Light Pollution Removal". Zoom in for better view.

Figure 13. Qualitative comparison of different methods. Specific tasks are marked in small grey text , "LES" means "Light Effects Suppression" and "LPR" means "Light Pollution Removal". Zoom in for better view.

Figure 14. Qualitative comparison of different methods. Specific tasks are marked in small grey text , "LES" means "Light Effects Suppression" and "LPR" means "Light Pollution Removal". Zoom in for better view.

Figure 15. Qualitative comparison of different methods. Specific tasks are marked in small grey text , "LES" means "Light Effects Suppression" and "LPR" means "Light Pollution Removal". Zoom in for better view.

Image Restoration

Image restoration uses computational methods to recover useful visual content from noisy, blurred, low-resolution, or otherwise degraded images.

See all 372 papers

IG-Diff: Complex Night Scene Restoration with Illumination-Guided Diffusion Model
May 14, 2026Yifan Chen, Fei Yin, Chunle Guo +2Diffusion Models for Image Restoration Diffusion Model Guidance
Tsinghua Univerisity · NanKai University
LUCID: Learning Unified Control for Image Deflaring and Exposure Mastery in Nighttime Photography
Jun 5, 2026Tingyu Yang, Yuan Cheng, Xiaoyun YuanDiffusion Models for Image Restoration Image Enhancement
Shanghai Jiao Tong University, China
CLIP-Guided Data Augmentation for Night-Time Image Dehazing
Apr 7, 2026Xining Ge, Weijun Yuan, Gengjia Chang +2Image Dehazing Data Augmentation
Hangzhou Dianzi University · Jinan University · Hefei University of Technology +2

May 14, 2026J/K move · Enter open · S save

IG-Diff: Complex Night Scene Restoration with Illumination-Guided Diffusion Model

Yifan Chen, Fei Yin, Chunle Guo, Chongyi Li+1

Tsinghua Univerisity · NanKai University

Diffusion Models for Image Restoration Diffusion Model Guidance Low-Light Image Enhancement All-in-One Image Restoration Image Restoration

Open paper AskPDF arXiv

In nighttime circumstances, it is challenging for individuals and machines to perceive their surroundings. While prevailing image restoration methods adeptly handle singular forms of degradation, they falter when confronted with intricate nocturnal scenes, such as the concurrent presence of weather and low-light conditions. Compounding this challenge, the lack of paired data that encapsulates the coexistence of low-light situations and other forms of degradation hinders the development of a comprehensive end-to-end solution. In this work, we contribute complex nighttime scene datasets that simulate both illumination degradation and other forms of deterioration. To address the complexity of night degradation, we propose an integration of an illumination-guided module embedded in the diffusion model to guide the illumination restoration process. Our model can preserve texture fidelity while contending with the adversities posed by various degradation in low-light scenarios.

Abstract

Image Restoration

Explore similar work

IG-Diff: Complex Night Scene Restoration with Illumination-Guided Diffusion Model

LUCID: Learning Unified Control for Image Deflaring and Exposure Mastery in Nighttime Photography

CLIP-Guided Data Augmentation for Night-Time Image Dehazing