In this paper, we address Novel Class Discovery (NCD), the task of unveiling new classes in a set of unlabeled samples given a labeled dataset with known classes. We exploit the peculiarities of NCD to build a new framework, named Neighborhood Contrastive Learning (NCL), to learn discriminative representations that are important to clustering performance. Our contribution is twofold. First, we find that a feature extractor trained on the labeled set generates representations in which a generic query sample and its neighbors are likely to share the same class. We exploit this observation to retrieve and aggregate pseudo-positive pairs with contrastive learning, thus encouraging the model to learn more discriminative representations. Second, we notice that most of the instances are easily discriminated by the network, contributing less to the contrastive loss. To overcome this issue, we propose to generate hard negatives by mixing labeled and unlabeled samples in the feature space. We experimentally demonstrate that these two ingredients significantly contribute to clustering performance and lead our model to outperform state-of-the-art methods by a large margin (e.g., clustering accuracy +13% on CIFAR-100 and +8% on ImageNet).

Neighborhood Contrastive Learning for Novel Class Discovery / Zhong, Zhun; Fini, Enrico; Roy, Subhankar; Luo, Zhiming; Ricci, Elisa; Sebe, Nicu. - (2021), pp. 10862-10870. (Intervento presentato al convegno CVPR 2021 tenutosi a virtual conference nel 19th-25th June 2021) [10.1109/CVPR46437.2021.01072].

Neighborhood Contrastive Learning for Novel Class Discovery

Zhong, Zhun;Fini, Enrico;Roy, Subhankar;Ricci, Elisa;Sebe, Nicu
2021-01-01

Abstract

In this paper, we address Novel Class Discovery (NCD), the task of unveiling new classes in a set of unlabeled samples given a labeled dataset with known classes. We exploit the peculiarities of NCD to build a new framework, named Neighborhood Contrastive Learning (NCL), to learn discriminative representations that are important to clustering performance. Our contribution is twofold. First, we find that a feature extractor trained on the labeled set generates representations in which a generic query sample and its neighbors are likely to share the same class. We exploit this observation to retrieve and aggregate pseudo-positive pairs with contrastive learning, thus encouraging the model to learn more discriminative representations. Second, we notice that most of the instances are easily discriminated by the network, contributing less to the contrastive loss. To overcome this issue, we propose to generate hard negatives by mixing labeled and unlabeled samples in the feature space. We experimentally demonstrate that these two ingredients significantly contribute to clustering performance and lead our model to outperform state-of-the-art methods by a large margin (e.g., clustering accuracy +13% on CIFAR-100 and +8% on ImageNet).
2021
Proceedings: 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition
Piscataway, NJ
IEEE
978-1-6654-4509-2
Zhong, Zhun; Fini, Enrico; Roy, Subhankar; Luo, Zhiming; Ricci, Elisa; Sebe, Nicu
Neighborhood Contrastive Learning for Novel Class Discovery / Zhong, Zhun; Fini, Enrico; Roy, Subhankar; Luo, Zhiming; Ricci, Elisa; Sebe, Nicu. - (2021), pp. 10862-10870. (Intervento presentato al convegno CVPR 2021 tenutosi a virtual conference nel 19th-25th June 2021) [10.1109/CVPR46437.2021.01072].
File in questo prodotto:
File Dimensione Formato  
Zhong_Neighborhood_Contrastive_Learning_for_Novel_Class_Discovery_CVPR_2021_paper.pdf

accesso aperto

Tipologia: Post-print referato (Refereed author’s manuscript)
Licenza: Tutti i diritti riservati (All rights reserved)
Dimensione 2.79 MB
Formato Adobe PDF
2.79 MB Adobe PDF Visualizza/Apri
Neighborhood_Contrastive_Learning_for_Novel_Class_Discovery.pdf

Solo gestori archivio

Tipologia: Versione editoriale (Publisher’s layout)
Licenza: Tutti i diritti riservati (All rights reserved)
Dimensione 1.86 MB
Formato Adobe PDF
1.86 MB Adobe PDF   Visualizza/Apri

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11572/326190
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 46
  • ???jsp.display-item.citation.isi??? 30
social impact