This paper focuses on improving the robustness of object detection systems under adverse weather conditions. To address this issue, which is essential for the advancement of autonomous driving technologies, we present a prompting methodology that generates adverse weather-based data augmentation using a diffusion model, Instruct Pix2Pix. The goal is to mitigate the adverse weather impact on the recognition ability of state-of-the-art object detection models, including Faster R-CNN and YOLOv10. We conduct experiments using the CARLA simulator and real image datasets, BDD100K and ACDC, to verify the effectiveness of the proposed data augmentation method in both simulated and real environments. The main contributions are to identify and quantify the performance differences of object detection models under adverse weather conditions and to show that a tailored data augmentation strategy improves the robustness of the models.
