Clear Nights Ahead: Towards Multi-Weather Nighttime Image Restoration

TL;DR

The paper tackles multi-weather nighttime image restoration by introducing AllWeatherNight, a large-scale illumination-aware dataset, and ClearNight, a unified framework that combines Retinex-based dual priors with a weather-aware dynamic specificity-commonality architecture. The approach models uneven illumination and weather degradations jointly, using a weather instructor and dynamic unit selection to adaptively handle diverse conditions. Extensive experiments show state-of-the-art performance on both synthetic and real-world nighttime images, with thorough ablations validating the necessity of the dataset and the effectiveness of the dual-prior and dynamic collaboration components. The work provides practical restoration capabilities and a scalable benchmark for real-world, complex nighttime environments.

Abstract

Restoring nighttime images affected by multiple adverse weather conditions is a practical yet under-explored research problem, as multiple weather conditions often 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 high-quality nighttime images with diverse compositional degradations, synthesized using our introduced illumination-aware degradation generation. Moreover, we present 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. In order to better represent the common and unique characters of multiple weather degradations, we introduce a weather-aware dynamic specific-commonality collaboration method, which identifies weather degradations and adaptively selects optimal candidate units associated with specific weather types. Our ClearNight achieves state-of-the-art performance on both synthetic and real-world images. Comprehensive ablation experiments validate the necessity of AllWeatherNight dataset as well as the effectiveness of ClearNight. Project Page: https://henlyta.github.io/ClearNight/

Figures (23)

• Figure 1: ClearNight is the first multi-weather nighttime image restoration framework, which effectively removes complex and coupled weather and flare degradations in one go.
• Figure 2: Real-world and synthetic samples in our AllWeatherNight dataset. The synthetic images effectively simulate real-world nighttime scenes with various degradations.
• Figure 3: Visualization of four synthesized variants of complex rain scene, where Weather Only and Flare Only denote synthesis with illumination-aware weather degradation and flare, respectively. Ours involves both degradations.
• Figure 4: Distribution of the our AllWeatherNight dataset.
• Figure 5: Overview of our ClearNight framework. ClearNight primarily comprises Retinex-based dual prior guidance as well as weather-aware dynamic specificity and commonality branches. The Retinex-based dual priors explicitly guide the network to focus on illumination regions and intrinsic textures. The weather-aware dynamic specificity branch adaptively accommodates various weather effects and collaborates with the commonality branch to effectively handle complex multi-weather scenes.