Self-supervised Contrastive Feature Refinement for Few-Shot Class-Incremental Learning

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

Few-Shot Class-Incremental Learning (FSCIL) is to learn novel classes with few data points incrementally, without forgetting old classes. It is very hard to capture the underlying patterns and traits of the few-shot classes. To meet the challenges, we propose a Self-supervised Contrastive Feature Refinement (SCFR) framework which tackles the FSCIL issue from three aspects. Firstly, we employ a self-supervised learning framework to make the network to learn richer representations and promote feature refinement. Meanwhile, we design virtual classes to improve the models robustness and generalization during training process. To prevent catastrophic forgetting, we attach Gaussian Noise to encountered prototypes to recall the distribution of known classes and maintain stability in the embedding space. SCFR offers a systematic solution which can effectively mitigate the issues of catastrophic forgetting and over-fitting. Experiments on widely recognized datasets, including CUB200, miniImageNet and CIFAR100, show remarkable performance than other mainstream works.

Original languageEnglish
Title of host publicationComputer-Aided Design and Computer Graphics - 18th International Conference, CAD/Graphics 2023, Proceedings
EditorsShi-Min Hu, Yiyu Cai, Paul Rosin
PublisherSpringer Science and Business Media Deutschland GmbH
Pages281-294
Number of pages14
ISBN (Print)9789819996650
DOIs
StatePublished - 2024
Event18th International Conference on Computer-Aided Design and Computer Graphics, CAD/Graphics 2023 - Shanghai, China
Duration: 19 Aug 202321 Aug 2023

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume14250 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference18th International Conference on Computer-Aided Design and Computer Graphics, CAD/Graphics 2023
Country/TerritoryChina
CityShanghai
Period19/08/2321/08/23

Keywords

  • Feature distribution recall
  • Few-shot class-incremental learning
  • Self-supervised learning
  • Virtual class augmentation

Fingerprint

Dive into the research topics of 'Self-supervised Contrastive Feature Refinement for Few-Shot Class-Incremental Learning'. Together they form a unique fingerprint.

Cite this