Clear Nights Ahead: Towards Multi-Weather Nighttime Image Restoration

¹Chongqing University of Posts and Telecommunications ²Zhejiang University

Abstract

Restoring nighttime images affected by multiple adverse weather conditions is a practical yet under-explored research problem, as multiple weather degradations usually coexist in the real world alongside various lighting effects at night. This paper first explores the challenging multi-weather nighttime image restoration task, where various types of weather degradations are intertwined with flare effects. To support the research, we contribute the AllWeatherNight dataset, featuring large-scale nighttime images with diverse compositional degradations. By employing illumination-aware degradation generation, our dataset significantly enhances the realism of synthetic degradations in nighttime scenes, providing a more reliable benchmark for model training and evaluation. Additionally, we propose ClearNight, a unified nighttime image restoration framework, which effectively removes complex degradations in one go. Specifically, ClearNight extracts Retinex-based dual priors and explicitly guides the network to focus on uneven illumination regions and intrinsic texture contents respectively, thereby enhancing restoration effectiveness in nighttime scenarios. Moreover, to more effectively model the common and unique characteristics of multiple weather degradations, ClearNight performs weather-aware dynamic specificity and commonality collaboration that adaptively allocates optimal sub-networks associated with specific weather types. Comprehensive experiments on both synthetic and real-world images demonstrate the necessity of the AllWeatherNight dataset and the superior performance of ClearNight.

AllWeatherNight Dataset

We observe that uneven lighting conditions in real-world nighttime scenes often interact with weather degradations. To synthesize more realistic nighttime images with adverse weather conditions, we introduce an illumination-aware degradation generation approach. To account for this, we derive Retinex decomposition to extract illumination maps as weights for subsequent weather degradation synthesis. Leveraging the proposed synthesis method, we put forward a dataset called AllWeatherNight. The generated images and labels in AllWeatherNight dataset is released under the BSD 3-Clause License.

@article{liu2025clearnight, title={Clear Nights Ahead: Towards Multi-Weather Nighttime Image Restoration}, author={Liu, Yuetong and Xu, Yunqiu and Wei, Yang and Bi, Xiuli and Xiao, Bin}, year={2025}, journal={arXiv preprint arXiv:2505.16479}, year={2025} }

Clear Nights Ahead: Towards Multi-Weather Nighttime Image Restoration

Abstract

AllWeatherNight Dataset

Distribution of the introduced AllWeatherNight dataset. We synthesize 8,000 nighttime images for model training, incorporating multi-degradation and single-degradation scenarios. The test dataset comprises two subsets----synthetic and real-world subset, each containing 1,000 nighttime images.

Dataset analysis. We create four distinct synthetic image variants of nighttime scenes to evaluate the effectiveness of our illumination-aware degradation generation using human ranking and NIQE. W and F indicate synthesis with illumination-aware weather degradation and flare effect respectively.

Visualization. Visualization of 4 different synthesized variants of adverse weather nighttime images, where weather only and flare only indicate synthesis with illumination-aware weather degradation and flare degradation respectively. Our method includes both degradation synthesis.

The t-SNE map of various nighttime image datasets. From left to right: feature distributions of the haze, rain, and snow nighttime datasets. The adverse weather feature distribution in our AllWeatherNight is closer to real-world nighttime scenes than existing nighttime synthetic datasets.

Dataset effectiveness on real-world data. Models trained on AllWeatherNight achieve superior performance on real-world images compared to those trained on the combination of existing nighttime datasets..

Method

BibTeX