Contents

Mitigating Embedding and Class Assignment Mismatch in Unsupervised Image Classification

Han, Sungwon / Park, Sungwon / Park, Sungkyu / Kim, Sundong / Cha, Meeyoung

  • 274 ITEM VIEW
  • 0 DOWNLOAD
Abstract

Unsupervised image classification is a challenging computer vision task. Deep learning-based algorithms have achieved superb results, where the latest approach adopts unified losses from embedding and class assignment processes. Since these processes inherently have different goals, jointly optimizing them may lead to a suboptimal solution. To address this limitation, we propose a novel two-stage algorithm in which an embedding module for pretraining precedes a refining module that concurrently performs embedding and class assignment. Our model outperforms SOTA when tested with multiple datasets, by substantially high accuracy of 81.0% for the CIFAR-10 dataset (i.e., increased by 19.3 percent points), 35.3% accuracy for CIFAR-100-20 (9.6 pp) and 66.5% accuracy for STL-10 (6.9 pp) in unsupervised tasks.

Issue Date
2020-08
Publisher
European Conference on Computer Vision
URI
https://archives.kdischool.ac.kr/handle/11125/54976
10.1007/978-3-030-58586-0_45
URL
https://link.springer.com/chapter/10.1007/978-3-030-58586-0_45
Conf. Name
16th European Conference on Computer Vision
Place
UK
Conference Date
2020-08-23
ISSN
0302-9743
Files in This Item:
    There are no files associated with this item.

Click the button and follow the links to connect to the full text. (KDI CL members only)

qrcode

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.

상단으로 이동