Harmonizing Light and Darkness: A Symphony of Prior-guided Data Synthesis and Adaptive Focus for Nighttime Flare Removal

TL;DR

This work tackles nighttime flare removal by introducing Flare7K*, a physically realistic, prior-guided data synthesis pipeline that enables multi-flare scenes whose brightness follows illumination laws. It pairs this dataset with an Adaptive Focus Module (AFM) that selectively masks non-flare regions, guiding restoration networks to focus on affected areas and preserve clean regions. Across multiple backbones, the combination yields state-of-the-art performance on real-world datasets, demonstrating improvements in both objective metrics and perceptual quality. The approach is flexible and can be integrated into existing flare-removal models, broadening practical impact for nighttime imaging and downstream tasks.

Abstract

Intense light sources often produce flares in captured images at night, which deteriorates the visual quality and negatively affects downstream applications. In order to train an effective flare removal network, a reliable dataset is essential. The mainstream flare removal datasets are semi-synthetic to reduce human labour, but these datasets do not cover typical scenarios involving multiple scattering flares. To tackle this issue, we synthesize a prior-guided dataset named Flare7K*, which contains multi-flare images where the brightness of flares adheres to the laws of illumination. Besides, flares tend to occupy localized regions of the image but existing networks perform flare removal on the entire image and sometimes modify clean areas incorrectly. Therefore, we propose a plug-and-play Adaptive Focus Module (AFM) that can adaptively mask the clean background areas and assist models in focusing on the regions severely affected by flares. Extensive experiments demonstrate that our data synthesis method can better simulate real-world scenes and several models equipped with AFM achieve state-of-the-art performance on the real-world test dataset.

Figures (8)

• Figure -1: (a) The schematic diagram
• Figure 0: (b) Real-world captured images
• Figure 2: Figure (a) shows the previous method for synthesizing flare images. Figure (b) is our proposed data synthesis pipeline. Our synthesis method can effectively simulate scenes with multiple flares of varying brightness.
• Figure 3: Overview of our training pipeline. The main purpose of AFM is to obtain the image that only contains the flare region and the network can be any commonly used image restoration network.
• Figure 4: Qualitative comparisons with the state-of-the-art methods on the real-world test dataset. Dai denotes the first Flare7Kdai2022flare7k work of his team and Dai++ denotes their second Flare7K++dai2023flare7k++ work.