A Survey on Real-Time Semantic Segmentation Based on Deep Learning

As modern life becomes increasingly intelligent, more and more applications require inferring relevant semantic data extracted from images for further analysis, such as network defense strategies, network security management, and so on. Due to the limited classification of network branches in existing review papers, this article introduces the commonly used single-branch structure, double-branch structure, and the latest three-branch structure in the field of real-time semantic segmentation; the serial and parallel structures of feature fusion modules. Finally, this article summarizes promising research directions in real-time semantic segmentation based on deep learning.