A Review of Attention Mechanisms in Computer Vision

As one of the popular and important techniques in deep learning, attention mechanism is widely used in various deep learning models of computer vision because of its excellent plug-and-play convenience. In this paper, the development process of attention mechanism is divided into four stages, reviewed and summarized in four aspects: channel attention, spatial attention, mixed channel and spatial attention, and self-attention, and its core ideas and key structures are introduced in detail. The current research status and future development direction of attention mechanism in image processing are summarized and explained.