Search papers, labs, and topics across Lattice.
This paper introduces OmniLight, a unified model for addressing adverse lighting conditions in images, using a Wavelet Domain Mixture-of-Experts (WD-MoE) architecture trained on multiple datasets. They contrast this with DINOLight, a specialized model trained on individual datasets, to study the impact of data distribution on model performance. Both OmniLight and DINOLight achieved top rankings in the NTIRE 2026 Challenge, demonstrating strong performance and generalization in lighting-related image restoration tasks.
A single model, OmniLight, can now handle diverse lighting conditions across multiple datasets as effectively as specialized models, thanks to a novel Wavelet Domain Mixture-of-Experts architecture.
Adverse lighting conditions, such as cast shadows and irregular illumination, pose significant challenges to computer vision systems by degrading visibility and color fidelity. Consequently, effective shadow removal and ALN are critical for restoring underlying image content, improving perceptual quality, and facilitating robust performance in downstream tasks. However, while achieving state-of-the-art results on specific benchmarks is a primary goal in image restoration challenges, real-world applications often demand robust models capable of handling diverse domains. To address this, we present a comprehensive study on lighting-related image restoration by exploring two contrasting strategies. We leverage a robust framework for ALN, DINOLight, as a specialized baseline to exploit the characteristics of each individual dataset, and extend it to OmniLight, a generalized alternative incorporating our proposed Wavelet Domain Mixture-of-Experts (WD-MoE) that is trained across all provided datasets. Through a comparative analysis of these two methods, we discuss the impact of data distribution on the performance of specialized and unified architectures in lighting-related image restoration. Notably, both approaches secured top-tier rankings across all three lighting-related tracks in the NTIRE 2026 Challenge, demonstrating their outstanding perceptual quality and generalization capabilities. Our codes are available at https://github.com/OBAKSA/Lighting-Restoration.
