Semantic Segmentation in Rainy Conditions Using Formula Driven Supervised Learning

In this study, we present a novel approach to enhance the performance of semantic segmentation in rainy conditions, a significant challenge in traffic monitoring. We transform artificial images generated by Formula Driven Supervised Learning (FDSL) to embed latent features of rainy conditions. By pre-training our model with these synthetic images, we aim to effectively capture the unique attributes of rainy conditions, thereby enhancing segmentation performance under such conditions. Our evaluation, conducted with the RaidaR open dataset, demonstrates that our method significantly enhances performance, even in the absence of real rainy images in the training dataset.