ECASeg: Enhancing Semantic Segmentation with Edge Context and Attention Strategy

Semantic segmentation has been a cornerstone for several visual perception-based applications, including autonomous driving, remote sensing and geographic information systems (RS-GIS), and medical diagnostics. By accurately segmenting the input visuals, an intelligent agent can perceive the environment and perform target operations appropriately. The emergence of deep learning (DL) has contributed to the development of cutting-edge segmentation models. However, there is scope for continual improvement and model optimization of existing solutions. Thus, this work pragmatically integrates an attention mechanism and edge features to improve the semantic segmentation. The exhaustive experimental analysis on the benchmark CamVid driving dataset show that the proposed approach achieves a mean Intersection over Union (mIoU) of 66.53%, which is a 5.12% improvement compared to the performance of a baseline model.