Presentation | 2022-12-16 Improving Pedestrian Attribute Recognition with Spatial Attention and Attribute Correlation Yichen Chen, Tetsuya Matsumoto, Yoshinori Takeuchi, Hiroaki Kudo, |
---|---|
PDF Download Page | PDF download Page Link |
Abstract(in Japanese) | (See Japanese page) |
Abstract(in English) | In the field of video surveillance, pedestrian attribute recognition is a focus technical thema of study. To predict certain attributes of a pedestrian, it is necessary to localize areas associated with such attributes from an image since region annotations for this task are unavailable. Previous methods try that they introduce heuristic body-part localization processes to improve local feature representations. However, they didn't use attributes to define local feature regions. In addition, correlations between attributes have not been utilized. In real-world video surveillance scenarios, visual pedestrian attributes such as gender, clothing color, and so on, are crucial for pedestrian attribute recognition. According to the correlation of attributes, it is considered useful to localize the regions. In this report, we propose an attribute estimation network from pedestrians based on spatial attention maps and attribute correlations. Spatial attention can help the network focus on where each attribute should be concerned, and the correlation between attributes helps to reduce some incorrect predictions. In the experiments using pedestrian attribute datasets, we obtained the results that the proposed method estimates binary attributes with roughly 90% accuracy and categorical attributes with around 75% accuracy. For the mA and F1 metrics, the Attribute correlation attention module can enhance performance by 1.36% and 1.57%, respectively. The Spatial attention module can also improve performance by 0.57% and 0.67%. |
Keyword(in Japanese) | (See Japanese page) |
Keyword(in English) | video surveillance / pedestrian attribute recognition / spatial attention / attributes correlation |
Paper # | IMQ2022-18 |
Date of Issue | 2022-12-09 (IMQ) |
Conference Information | |
Committee | IMQ |
---|---|
Conference Date | 2022/12/16(1days) |
Place (in Japanese) | (See Japanese page) |
Place (in English) | Nishi-Chiba Campus, Chiba Univ. |
Topics (in Japanese) | (See Japanese page) |
Topics (in English) | Image Media Quality, etc. |
Chair | Kenya Uomori(Osaka Univ.) |
Vice Chair | Mitsuru Maeda(Canon) |
Secretary | Mitsuru Maeda(Nagoya Univ.) |
Assistant | Masato Tsukada(Univ. of Tsukuba) / Takashi Yamazoe(Seikei Univ.) |
Paper Information | |
Registration To | Technical Committee on Image Media Quality |
---|---|
Language | ENG-JTITLE |
Title (in Japanese) | (See Japanese page) |
Sub Title (in Japanese) | (See Japanese page) |
Title (in English) | Improving Pedestrian Attribute Recognition with Spatial Attention and Attribute Correlation |
Sub Title (in English) | |
Keyword(1) | video surveillance |
Keyword(2) | pedestrian attribute recognition |
Keyword(3) | spatial attention |
Keyword(4) | attributes correlation |
1st Author's Name | Yichen Chen |
1st Author's Affiliation | Nagoya University(Nagoya Univ.) |
2nd Author's Name | Tetsuya Matsumoto |
2nd Author's Affiliation | Nagoya University(Nagoya Univ.) |
3rd Author's Name | Yoshinori Takeuchi |
3rd Author's Affiliation | Daido University(Daido Univ.) |
4th Author's Name | Hiroaki Kudo |
4th Author's Affiliation | Nagoya University(Nagoya Univ.) |
Date | 2022-12-16 |
Paper # | IMQ2022-18 |
Volume (vol) | vol.122 |
Number (no) | IMQ-317 |
Page | pp.pp.16-21(IMQ), |
#Pages | 6 |
Date of Issue | 2022-12-09 (IMQ) |