Prototype calibration for long tailed recognition

doi:10.1109/ICME55011.2023.00363

	Prototype calibration for long tailed recognition
	Zhongan Wang; Shuai Shi; Yingna Wu; Rui Yang
	2023-07-10
会议录名称	2023 IEEE INTERNATIONAL CONFERENCE ON MULTIMEDIA AND EXPO (ICME)
ISSN	1945-7871
卷号	2023-July
页码	2123-2128
发表状态	已发表
DOI	10.1109/ICME55011.2023.00363
摘要	In the real world, data distribution usually presents imbalanced characteristics, such as long-tailed distribution, which is generally divided into the head and tail classes. For tail classes, image features cannot be represented well due to insufficient training samples. It is a vital task to learn discriminative image representation on imbalanced data distribution. In our work, through exploring prototype information, we propose a prototype-based contrastive learning(PCL) loss and prototype-based feature augmentation(PFA) module to improve the accuracy of the classifier on the imbalanced dataset. Specifically, we utilize the classifier parameters to generate learnable embeddings, which can be regarded as the class centers after using metric learning. The PFA module generates the image features of each tail class with the help of head class information. We validate our approach on common long-tailed benchmarks. Our results indicate that the PCL and PFA make the classification model achieve significant performance boosts on these benchmarks. © 2023 IEEE.
会议录编者/会议主办者	et al. ; IEEE Circuits and Systems (CAS) ; IEEE Communications (ComSoc) ; IEEE Computer (CS) ; IEEE Signal Processing (SPS) ; Tencent
关键词	Long tailed recognition Few-shot learning Prototype Network Contrastive Learning
会议名称	2023 IEEE International Conference on Multimedia and Expo, ICME 2023
会议地点	Brisbane, Australia
会议日期	10-14 July 2023
URL	查看原文
收录类别	EI
语种	英语
出版者	IEEE Computer Society
EI入藏号	20233814738670
EI主题词	Image representation
EISSN	1945-788X
EI分类号	716.1 Information Theory and Signal Processing ; 723.4 Artificial Intelligence ; 723.5 Computer Applications ; 741.2 Vision ; 903.1 Information Sources and Analysis
原始文献类型	Conference article (CA)
来源库	IEEE
文献类型	会议论文
条目标识符	https://kms.shanghaitech.edu.cn/handle/2MSLDSTB/333423
专题	创意与艺术学院_PI研究组(P)_武颖娜组物质科学与技术学院_硕士生信息科学与技术学院_硕士生创意与艺术学院_PI研究组(P)_杨锐组
作者单位	ShanghaiTech University, Shanghai, China
第一作者单位	上海科技大学
第一作者的第一单位	上海科技大学
推荐引用方式 GB/T 7714	Zhongan Wang,Shuai Shi,Yingna Wu,et al. Prototype calibration for long tailed recognition[C]//et al., IEEE Circuits and Systems (CAS), IEEE Communications (ComSoc), IEEE Computer (CS), IEEE Signal Processing (SPS), Tencent:IEEE Computer Society,2023:2123-2128.