Encoder-Decoder based Segmentation Model for UAV Street Scene Images

Abstract

Global contextual information needs to be modeled precisely for accurate segmentation of images taken by Unmanned Aerial Vehicles (UAVs). This paper presents a transformer-based method for UAV street scene semantic segmentation. The method uses an encoder-decoder-based architecture to capture local and global context information in UAV images. Experimental result of the proposed method shows competitive performance against state-of-the-art methods by achieving mIoU of 61.93% on UAVid dataset.