Cvpr2022 𞓜 Heidelberg University's course "deep visual similarity and measurement learning"

CVPR 2022 Offline meeting will be held on 2022 year 6 month 21 Japan -24 It was held in New Orleans, USA . This year, the number of submissions reached a new high of more than 10000 , among 2067 Papers were received . Scholars have brought a series of tutorials . From Heidelberg University 《 Deep visual similarity and metric learning 》 course ,200+ page ppt Worthy of attention .

Learning visual similarity is very important for various visual tasks , Such as image clustering 、 Face detection or image retrieval , So as to lay the foundation for daily application , Such as the smart arrangement of image sets on smart phones 、 Web scale image search in the browser or recommend products for online shopping . today , The main method of learning to capture the visual representation of similarity is depth measurement learning , It is specifically aimed at the novel 、 Retrieving objects and images from invisible classes . Besides , Similarity learning is closely related to contrast learning , Comparative learning is the dominant method of self - supervised learning , They are transfer learning .

In this tutorial , We will go into depth measurement learning (DML) Leading learning paradigm , And how to actually evaluate their ( Out of distribution ) The future direction of generalization . say concretely , This tutorial will cover the following topics : (i) DML Overview of the objective function ,(ii) Advanced and context sensitive DML The formula ,(iii) DML The importance of data sampling in ,(iv) Fair and realistic assessment DML Best practices for methods , Last ,(v) We will DML Associated with the related fields of computer vision and pattern recognition , Such as comparative learning , Study with fewer samples , Transfer learning and face recognition .

Address ：https://dvsml2022-tutorial.github.io/program.html