分割
GSM演进的增强数据速率
计算机科学
背景(考古学)
人工智能
自然语言处理
地理
考古
作者
Kaili Yu,Zhenxue Chen,Mengting Ye,Yixin Guo,Q. M. Jonathan Wu
摘要
Semantic segmentation is widely needed in many application areas. Driven by the development of deep learning, semantic segmentation methods have made significant progress in accuracy, robustness, and speed. Based on the short-term dense concatenate network, this paper proposes an edge-guided context real-time semantic segmentation network (EGCNet), which further improves the detail branch, extracts edge information from feature maps, and enhances the network’s ability to extract edge information through edge-assisted training. Because the top-level feature map contains minimal detailed information, the network only extracts the boundary of the low-level feature map such that the useful information is utilized as much as possible while reducing the computational load. In addition, we improve the attention-refining module by embedding position information, allowing it to capture the target structure more accurately. EGCNet achieves 71 FPS and 78.6\% mIoU on the Cityscapes dataset at a low computational cost, making it an efficient real-time semantic segmentation network.
科研通智能强力驱动
Strongly Powered by AbleSci AI