Paper Abstract and Keywords |
Presentation |
2021-12-03 11:20
Open Set Recognition Focusing on Multidimensional Feature Space Daiju Kanaoka (Kyutech), Yuichiro Tanaka, Hakaru Tamukoh (Kyutech/Kyushu Institute of Technology/Research Center for Neuro) SIS2021-23 |
Abstract |
(in Japanese) |
(See Japanese page) |
(in English) |
In general, image recognition assumes that all classes used in testing are known. Therefore, when an unknown class is input, it cannot be recognized as unknown. Methods that make this possible are called open set recognition. In this study, we propose an open set recognition method focusing on a multidimensional feature space of the recognition model. The experimental results show that the macro-F1 at MNIST is 0.838, which is better than the state-of-the-art method. We find the potential of a method focusing on multidimensional feature spaces. |
Keyword |
(in Japanese) |
(See Japanese page) |
(in English) |
open set recognition / deep learning / unknown class / / / / / |
Reference Info. |
IEICE Tech. Rep., vol. 121, no. 284, SIS2021-23, pp. 11-14, Dec. 2021. |
Paper # |
SIS2021-23 |
Date of Issue |
2021-11-26 (SIS) |
ISSN |
Online edition: ISSN 2432-6380 |
Copyright and reproduction |
All rights are reserved and no part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without permission in writing from the publisher. Notwithstanding, instructors are permitted to photocopy isolated articles for noncommercial classroom use without fee. (License No.: 10GA0019/12GB0052/13GB0056/17GB0034/18GB0034) |
Download PDF |
SIS2021-23 |
|