DEANet: A Real-Time Image Semantic Segmentation Method Based on Dual Efficient Attention Mechanism

Image semantic segmentation is the basis of performing various tasks in computer vision. It has been widely used in medical imaging, robotics and many other fields. However, the existing image semantic segmentation technology cannot improve the segmentation speed while ensuring the segmentation accuracy, and cannot meet the requirements of real-time applications. Therefore, this paper proposes a real-time image semantic segmentation method based on dual efficient attention mechanism (DEANet). Pyramid sampling is introduced into the channel dimension to extract multi-scale information, and higher resolution aggregation features are adopted as the input of the spatial dimension. It can achieve high efficiency and accuracy of image semantic segmentation. The proposed DEANet was tested on two classic datasets. On the Cityscapes dataset, when the input size is 512 x 1024, the segmentation accuracy reaches 74.90% mIoU, and the segmentation speed reaches 99.91FPS. On the CamVid dataset, when the input size is 360 x 480, the segmentation accuracy reaches 70.07% mIoU and the segmentation speed reaches 142.72 FPS.