An overview of the latest research progress of "efficient deep segmentation of labels" at Shanghai Jiaotong University, which comprehensively expounds the deep segmentation methods of unsupervised, ro

Thesis link ：https://arxiv.org/pdf/2207.01223.pdf

With the rapid development of deep learning , Segmentation technology, one of the basic tasks of computer vision, has made great progress . However , Current segmentation algorithms mainly rely on the availability of pixel level annotation , This is usually expensive 、 Cumbersome and laborious . To lighten the burden , In the past few years , People pay more and more attention to the establishment of efficient labels 、 Segmentation algorithm based on deep learning . This paper gives a comprehensive overview of efficient label segmentation methods . So , We will start with different types of weak tags ( Including unsupervised 、 Rough supervision 、 Incomplete supervision and Noise Supervision ) Supervision provided , And supplemented by the type of segmentation problem ( Including semantic segmentation 、 Instance segmentation and panoramic segmentation ), A taxonomy has been developed to organize these methods . Next , We summarize the existing efficient label segmentation methods from a unified perspective , An important issue was discussed : How to bridge the gap between weak supervision and intensive prediction —— Most of the current methods are based on heuristic Apriori , Such as cross pixel similarity 、 Cross label constraints 、 Cross view consistency 、 Cross image relationships, etc . Last , We put forward our own views on the future research direction of efficient deep segmentation of tags .