IAF-Net: Illumination-Adaptive Fusion for Low-Light Urban Road Segmentation

2026-05-29

AI Summary

This work addresses the significant performance degradation of urban road semantic segmentation under low-light conditions, a challenge exacerbated by the lack of illumination-aware modeling of modality reliability in existing methods. To this end, we propose IAF-Net, an end-to-end framework that explicitly models the impact of illumination on the reliability of RGB and geometric modalities. The framework incorporates an illumination-adaptive fusion module to dynamically adjust fusion weights and introduces a luminance-modulated attention decoder to enhance feature selection in low-light scenarios. To facilitate research in this domain, we construct two novel low-light, multi-weather datasets: nuScenes-NRS and CARLA-MWRS. Experimental results demonstrate that IAF-Net achieves state-of-the-art performance on nuScenes-NRS, with the IAF module alone contributing up to a 0.70% improvement in MaxF, while also exhibiting robustness across adverse weather conditions on CARLA-MWRS.

Abstract

Semantic road segmentation is important for autonomous driving, but existing methods suffer severe performance degradation under low-light conditions. Many existing multi-modal fusion methods do not explicitly adapt to illumination-dependent changes in modality reliability, which can propagate degraded RGB features into the fused representation at night. We propose IAF-Net (Illumination-Adaptive Fusion Network), an end-to-end framework with illumination-adaptive fusion for robust road segmentation across different lighting conditions. It dynamically adjusts fusion weights of RGB and geometric features via the core Illumination-Adaptive Fusion (IAF) module, and enhances low-light feature selection with a brightness-modulated attention decoder. We also construct two dedicated datasets: nuScenes Nighttime Road Segmentation (nuScenes-NRS) and CARLA Multi-Weather Road Segmentation (CARLA-MWRS). Experiments on nuScenes-NRS show state-of-the-art overall performance among the compared methods, while CARLA-MWRS further validates robustness across adverse weather conditions. Ablation studies on a 40% training subset further highlight the importance of the IAF module, which provides the largest individual gain of 0.70% in MaxF.

Problem

• low-light segmentation
• modality reliability
• illumination adaptation
• road segmentation
• autonomous driving

Innovation

• Illumination-Adaptive Fusion
• Low-Light Segmentation
• Multi-Modal Fusion
• Brightness-Modulated Attention
• Urban Road Segmentation