Presentation | 2021-03-04 Evaluations on Two-stream Feature-fusion Architecture with Shifting-subclip and Hard Positive Mining for Person Re-identification Yuki Hiroi, Wataru Kameyama, |
---|---|
PDF Download Page | PDF download Page Link |
Abstract(in Japanese) | (See Japanese page) |
Abstract(in English) | As a person re-identification technology using spatial-temporal features (video-based Re-ID), we have proposed two-stream feature-fusion architecture which parallelly extracts entire body-image features and partial body-image features, and shifting-subclip which is to apply CNN several times to the neighboring frames by shifting frames. We have also proposed hard positive mining for Re-ID to improve the identification accuracy of persons who are difficult to identify. However, these proposed methods are individually evaluated only using MARS dataset. Therefore, in this paper, we report the results of all combined methods compared with the conventional method in multiple datasets. As the result of the experiment in ILIDS-VID, PRID and MARS datasets, although the effectiveness of partial body-image features are different depending on the appearance of persons in each dataset, we confirm that the proposed two-stream feature-fusion architecture is effective in these three datasets. We also confirm that the proposed shifting-subclip is effective only for datasets with a small number of frames. Due to the nature of the datasets, the proposed hard positive mining is evaluated only using MARS dataset, however, the accuracy is further improved by combining it with the two-stream feature-fusion architecture and the shifting-subclip. Therefore, it is suggested that the proposed three methods are effective for various datasets in general. |
Keyword(in Japanese) | (See Japanese page) |
Keyword(in English) | Person Re-identification / Two-stream Feature-fusion Architecture / Shifting-subclip / Hard Positive Mining |
Paper # | PRMU2020-77 |
Date of Issue | 2021-02-25 (PRMU) |
Conference Information | |
Committee | PRMU / IPSJ-CVIM |
---|---|
Conference Date | 2021/3/4(2days) |
Place (in Japanese) | (See Japanese page) |
Place (in English) | Online |
Topics (in Japanese) | (See Japanese page) |
Topics (in English) | Computer Vision and Pattern Recognition for specific environment |
Chair | Yoichi Sato(Univ. of Tokyo) |
Vice Chair | Akisato Kimura(NTT) / Masakazu Iwamura(Osaka Pref. Univ.) |
Secretary | Akisato Kimura(Mobility Technologies) / Masakazu Iwamura(Chubu Univ.) |
Assistant | Takashi Shibata(NTT) / Masashi Nishiyama(Tottori Univ.) |
Paper Information | |
Registration To | Technical Committee on Pattern Recognition and Media Understanding / 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) | Evaluations on Two-stream Feature-fusion Architecture with Shifting-subclip and Hard Positive Mining for Person Re-identification |
Sub Title (in English) | |
Keyword(1) | Person Re-identification |
Keyword(2) | Two-stream Feature-fusion Architecture |
Keyword(3) | Shifting-subclip |
Keyword(4) | Hard Positive Mining |
1st Author's Name | Yuki Hiroi |
1st Author's Affiliation | Waseda University(Waseda Univ.) |
2nd Author's Name | Wataru Kameyama |
2nd Author's Affiliation | Waseda University(Waseda Univ.) |
Date | 2021-03-04 |
Paper # | PRMU2020-77 |
Volume (vol) | vol.120 |
Number (no) | PRMU-409 |
Page | pp.pp.47-52(PRMU), |
#Pages | 6 |
Date of Issue | 2021-02-25 (PRMU) |