Presentation | 2019-09-05 A method for visualizing the cause of misrecognition in object recognition using CNN Tomonori Kubota, Yasuyuki Murata, Yoshifumi Uehara, Akira Nakagawa, |
---|---|
PDF Download Page | PDF download Page Link |
Abstract(in Japanese) | (See Japanese page) |
Abstract(in English) | In this paper, we propose a method for visualizing the cause of misrecognition in object recognition using CNN. By this method, it becomes possible to extract and visualize at pixel grain size the place of cause of the image which deteriorates classification probability (Score) of correct answer class in the misrecognition image. And, it is possible to correct the extracted information to the image in which the classification probability of the correct answer class is improved by affecting the misrecognition image, and it can be confirmed that the extracted information correctly shows the cause of the misrecognition. This time, this paper shows the effectiveness of this method by a pre-trained model for discriminating a “car name and model year”. |
Keyword(in Japanese) | (See Japanese page) |
Keyword(in English) | object recognition / convolutional neural network / inference / misrecognition / visualizing / XAI |
Paper # | PRMU2019-25,MI2019-44 |
Date of Issue | 2019-08-28 (PRMU, MI) |
Conference Information | |
Committee | PRMU / MI / IPSJ-CVIM |
---|---|
Conference Date | 2019/9/4(2days) |
Place (in Japanese) | (See Japanese page) |
Place (in English) | |
Topics (in Japanese) | (See Japanese page) |
Topics (in English) | |
Chair | Yoichi Sato(Univ. of Tokyo) / Yoshiki Kawata(Tokushima Univ.) |
Vice Chair | Toru Tamaki(Hiroshima Univ.) / Akisato Kimura(NTT) / Takayuki Kitasaka(Aichi Inst. of Tech.) / Hidekata Hontani(Nagoya Inst. of Tech.) |
Secretary | Toru Tamaki(NTT) / Akisato Kimura(OMRON SINICX) / Takayuki Kitasaka(Yamaguchi Univ.) / Hidekata Hontani(Univ. of Hyogo) |
Assistant | Yusuke Uchida(DeNA) / Takayoshi Yamashita(Chubu Univ.) / Hotaka Takizawa(Tsukuba Univ.) / Yoshito Otake(NAIST) |
Paper Information | |
Registration To | Technical Committee on Pattern Recognition and Media Understanding / Technical Committee on Medical Imaging / Special Interest Group on Computer Vision and Image Media |
---|---|
Language | JPN |
Title (in Japanese) | (See Japanese page) |
Sub Title (in Japanese) | (See Japanese page) |
Title (in English) | A method for visualizing the cause of misrecognition in object recognition using CNN |
Sub Title (in English) | |
Keyword(1) | object recognition |
Keyword(2) | convolutional neural network |
Keyword(3) | inference |
Keyword(4) | misrecognition |
Keyword(5) | visualizing |
Keyword(6) | XAI |
1st Author's Name | Tomonori Kubota |
1st Author's Affiliation | Fujitsu Laboratories LTD.(Fujitsu Lab.) |
2nd Author's Name | Yasuyuki Murata |
2nd Author's Affiliation | Fujitsu Software Technologies Limited(FST) |
3rd Author's Name | Yoshifumi Uehara |
3rd Author's Affiliation | Fujitsu Laboratories LTD.(Fujitsu Lab.) |
4th Author's Name | Akira Nakagawa |
4th Author's Affiliation | Fujitsu Laboratories LTD.(Fujitsu Lab.) |
Date | 2019-09-05 |
Paper # | PRMU2019-25,MI2019-44 |
Volume (vol) | vol.119 |
Number (no) | PRMU-192,MI-193 |
Page | pp.pp.99-104(PRMU), pp.99-104(MI), |
#Pages | 6 |
Date of Issue | 2019-08-28 (PRMU, MI) |