Is CNN Really Looking at Your Face?

Hiroya Kawai, Takashi Kozu, Koichi Ito, Hwann Tzong Chen, Takafumi Aoki

研究成果: 書籍の章/レポート/Proceedings会議への寄与査読

1 被引用数 (Scopus)

抄録

Face recognition has dramatically improved its performance with the advent of deep learning, especially Convolutional Neural Networks (CNNs), while they have raised a new issue: the difficulty in interpreting the results. The question is what CNN looks at in a face image to identify a person. To answer this question, this paper presents a simple and novel analysis of deep face recognition based on facial parts. We evaluate the recognition accuracy of face images with specific regions masked using face segmentation labels. Our analysis clarifies what CNNs really need in face images for face recognition. The paper concludes with an application of face recognition models to general visualization methods and the problems contained in some classical face image datasets.

本文言語英語
ホスト出版物のタイトルPattern Recognition - 6th Asian Conference, ACPR 2021, Revised Selected Papers
編集者Christian Wallraven, Qingshan Liu, Hajime Nagahara
出版社Springer Science and Business Media Deutschland GmbH
ページ525-539
ページ数15
ISBN(印刷版)9783031023743
DOI
出版ステータス出版済み - 2022
イベント6th Asian Conference on Pattern Recognition, ACPR 2021 - Virtual, Online
継続期間: 2021 11月 92021 11月 12

出版物シリーズ

名前Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
13188 LNCS
ISSN(印刷版)0302-9743
ISSN(電子版)1611-3349

会議

会議6th Asian Conference on Pattern Recognition, ACPR 2021
CityVirtual, Online
Period21/11/921/11/12

フィンガープリント

「Is CNN Really Looking at Your Face?」の研究トピックを掘り下げます。これらがまとまってユニークなフィンガープリントを構成します。

引用スタイル