Swin-APT: An Enhancing Swin-Transformer Adaptor for Intelligent Transportation

Abstract

Artificial Intelligence has been widely applied in intelligent transportation systems. In this work, Swin-APT, a deep learning-based approach for semantic segmentation and object detection in intelligent transportation systems is presented. Swin-APT includes a lightweight network and a multiscale adapter network designed for image semantic segmentation and object detection tasks. An inter-frame consistency module is proposed to extract more accurate road information from images. Experimental results on four datasets: BDD100K, CamVid, SYNTHIA, and CeyMo, demonstrate that Swin-APT outperforms the baseline by 13.1%. Furthermore, experiments on the road marking detection benchmark show an improvement of 1.85% of mAcc.