带噪声标签细粒度图像分类方法研究综述

Review on Fine-Grained Image Classification with Noise Labels

要:开展带噪声标签细粒度图像分类方法综述,在细粒度图像分类问题定义及方法类型梳理的基础上,对噪声标签问题进行描述,总结了标签噪声的来源,从影响因素依赖关系和实例依赖关系两个角度对标签噪声进行分类,分析了标签噪声对细粒度图像分类任务的影响,从基于显式噪声模型方法、基于隐式噪声模型方法两个维度梳理了细粒度图像分类的研究进展,提出了未来值得关注的研究主题。

A review of classification methods of fine-grained images with noise labels was conducted.Firstly,based on the definition of fine-grained image classification problems and the classification of method types,the problems of noise labels were described,as well as the sourc-es of label noise were summarized.Then,label noise was classified from the perspectives of in-fluence factor dependency and instance dependency,and the impact of label noise on fine-grained image classification was analyzed.Furthermore,the progress of fine-grained image classification was discussed from two dimensions:explicit noise model based methods and implicit noise model based methods.Thereby,future noteworthy research topics were proposed.