Single-stage instance segmentation:a review of network structure research from local to global
Single-stage instance segmentation is a hot research topic in the field of deep learning in recent years,in which the instance-level segmentation of images is realized by paralleling methods of object detection and object segmentation.This method has been widely used in the field of image object segmentation.Firstly,the basic principle of single-stage instance segmentation is described.Secondly,the network structure of single-stage instance segmentation is sorted from local and overall aspects.In terms of local network structure,the summarization includes three aspects:feature extraction,feature fusion,and feature prediction.Specifi-cally for the feature prediction,the generation method of the object boundary frame is classified according to the idea of anchor frame to non-anchor frame.The representation of object mask is classified according to the idea of global mask to local mask.The global mask methods include prototype coefficient method,object position method,and object boundary method,while the local mask methods include object contour method,object position method,and object feature method.In terms of the overall network struc-ture,the 22 mainstream network structures are summarized.Then,the development status of single-stage instance segmentation in medical image segmentation,video image segmentation,remote sensing image segmentation,and other application fields are sum-marized.Finally,the development directions of single-stage instance segmentation are prospected.
NETL
NSTL
万方数据
