RHRSegNet: Relighting High-Resolution Night-Time Semantic Segmentation

TL;DR

RHRSegNet is a novel deep learning model that enhances nighttime semantic segmentation by combining relighting techniques with high-resolution networks, significantly improving performance in low-light conditions for autonomous driving.

Contribution

It introduces a relighting model integrated with HRNet for better nighttime scene segmentation, addressing challenges of low-light and complex lighting conditions.

Findings

Achieves 5% improvement in segmentation accuracy on nighttime datasets.

Effectively handles complex lighting and shadow effects in low-light scenes.

Demonstrates robustness across multiple large-scale nighttime datasets.

Abstract

Night time semantic segmentation is a crucial task in computer vision, focusing on accurately classifying and segmenting objects in low-light conditions. Unlike daytime techniques, which often perform worse in nighttime scenes, it is essential for autonomous driving due to insufficient lighting, low illumination, dynamic lighting, shadow effects, and reduced contrast. We propose RHRSegNet, implementing a relighting model over a High-Resolution Network for semantic segmentation. RHRSegNet implements residual convolutional feature learning to handle complex lighting conditions. Our model then feeds the lightened scene feature maps into a high-resolution network for scene segmentation. The network consists of a convolutional producing feature maps with varying resolutions, achieving different levels of resolution through down-sampling and up-sampling. Large nighttime datasets are used for training and evaluation, such as NightCity, City-Scape, and Dark-Zurich datasets. Our proposed model increases the HRnet segmentation performance by 5% in low-light or nighttime images.