# SemOD: Semantic-Enabled Object Detection Network Under Various Weather Conditions

## TL;DR

This paper introduces a new object detection network that uses semantic information to improve performance in various weather conditions for autonomous driving.

## Contribution

The novel contribution is a semantic-enabled network that enhances object detection by integrating semantic guidance into image restoration and detection.

## Key findings

- The proposed method achieves mAP improvements of 1.49% to 8.78% over existing approaches.
- Semantic information improves visual coherence and object boundary delineation in degraded images.
- The network includes a preprocessing unit and detection unit for robust performance in diverse weather.

## Abstract

In the field of autonomous driving, camera-based perception models are mostly trained on clear weather data. Models designed to handle specific weather conditions often lack generalization to dynamically changing environments and primarily focus on weather removal rather than robust perception. This paper proposes a semantic-enabled network for object detection under diverse weather conditions. Semantic information enables the model to generate plausible content in missing regions and accurately delineate object boundaries. It also preserves visual coherence and realism across both restored and original image areas, thereby facilitating image transformation and object recognition. Specifically, our architecture consists of a Preprocessing Unit (PPU) and a Detection Unit (DTU), where the PPU utilizes a U-shaped network enriched with semantics to refine degraded images, and the DTU integrates this semantic information for object detection using a modified YOLO network. Extensive experiments demonstrate that the proposed method achieves mAP improvements ranging from 1.49% to 8.78% compared with existing approaches across multiple benchmark datasets under various weather conditions. These results demonstrate the effectiveness of semantic guidance in image enhancement and object detection, providing a comprehensive framework for improving detection performance. The source code will be made publicly available.