Self-supervised recalibration network for person re-identification
The attention mechanism can extract salient features in images, which has been proved to be effective in improving the performance of person re-identification (Re-ID). However, most of the existing attention modules have the following two shortcomings: On the one hand, they mostly use global average...
| Main Authors: | , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
KeAi Communications Co., Ltd.
2024-01-01
|
| Series: | Defence Technology |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S2214914723000119 |
| _version_ | 1797334819805331456 |
|---|---|
| author | Shaoqi Hou Zhiming Wang Zhihua Dong Ye Li Zhiguo Wang Guangqiang Yin Xinzhong Wang |
| author_facet | Shaoqi Hou Zhiming Wang Zhihua Dong Ye Li Zhiguo Wang Guangqiang Yin Xinzhong Wang |
| author_sort | Shaoqi Hou |
| collection | DOAJ |
| description | The attention mechanism can extract salient features in images, which has been proved to be effective in improving the performance of person re-identification (Re-ID). However, most of the existing attention modules have the following two shortcomings: On the one hand, they mostly use global average pooling to generate context descriptors, without highlighting the guiding role of salient information on descriptor generation, resulting in insufficient ability of the final generated attention mask representation; On the other hand, the design of most attention modules is complicated, which greatly increases the computational cost of the model. To solve these problems, this paper proposes an attention module called self-supervised recalibration (SR) block, which introduces both global and local information through adaptive weighted fusion to generate a more refined attention mask. In particular, a special ''Squeeze-Excitation'' (SE) unit is designed in the SR block to further process the generated intermediate masks, both for nonlinearizations of the features and for constraint of the resulting computation by controlling the number of channels. Furthermore, we combine the most commonly used ResNet-50 to construct the instantiation model of the SR block, and verify its effectiveness on multiple Re-ID datasets, especially the mean Average Precision (mAP) on the Occluded-Duke dataset exceeds the state-of-the-art (SOTA) algorithm by 4.49%. |
| first_indexed | 2024-03-08T08:26:45Z |
| format | Article |
| id | doaj.art-b8f76792937f4956a14689a11cddafc7 |
| institution | Directory Open Access Journal |
| issn | 2214-9147 |
| language | English |
| last_indexed | 2024-03-08T08:26:45Z |
| publishDate | 2024-01-01 |
| publisher | KeAi Communications Co., Ltd. |
| record_format | Article |
| series | Defence Technology |
| spelling | doaj.art-b8f76792937f4956a14689a11cddafc72024-02-02T04:39:08ZengKeAi Communications Co., Ltd.Defence Technology2214-91472024-01-0131163178Self-supervised recalibration network for person re-identificationShaoqi Hou0Zhiming Wang1Zhihua Dong2Ye Li3Zhiguo Wang4Guangqiang Yin5Xinzhong Wang6Shenzhen Institute of Information Technology, Shenzhen, 518172, China; University of Electronic Science and Technology of China, Chengdu, 611731, China; Kash Institute of Electronics and Information Industry, Kashi, 844099, ChinaUniversity of Electronic Science and Technology of China, Chengdu, 611731, ChinaUniversity of Electronic Science and Technology of China, Chengdu, 611731, ChinaShenzhen Institute of Information Technology, Shenzhen, 518172, China; University of Electronic Science and Technology of China, Chengdu, 611731, China; Kash Institute of Electronics and Information Industry, Kashi, 844099, ChinaUniversity of Electronic Science and Technology of China, Chengdu, 611731, China; Corresponding author.University of Electronic Science and Technology of China, Chengdu, 611731, China; Corresponding author.Shenzhen Institute of Information Technology, Shenzhen, 518172, China; Corresponding author.The attention mechanism can extract salient features in images, which has been proved to be effective in improving the performance of person re-identification (Re-ID). However, most of the existing attention modules have the following two shortcomings: On the one hand, they mostly use global average pooling to generate context descriptors, without highlighting the guiding role of salient information on descriptor generation, resulting in insufficient ability of the final generated attention mask representation; On the other hand, the design of most attention modules is complicated, which greatly increases the computational cost of the model. To solve these problems, this paper proposes an attention module called self-supervised recalibration (SR) block, which introduces both global and local information through adaptive weighted fusion to generate a more refined attention mask. In particular, a special ''Squeeze-Excitation'' (SE) unit is designed in the SR block to further process the generated intermediate masks, both for nonlinearizations of the features and for constraint of the resulting computation by controlling the number of channels. Furthermore, we combine the most commonly used ResNet-50 to construct the instantiation model of the SR block, and verify its effectiveness on multiple Re-ID datasets, especially the mean Average Precision (mAP) on the Occluded-Duke dataset exceeds the state-of-the-art (SOTA) algorithm by 4.49%.http://www.sciencedirect.com/science/article/pii/S2214914723000119Person re-identificationAttention mechanismGlobal informationLocal informationAdaptive weighted fusion |
| spellingShingle | Shaoqi Hou Zhiming Wang Zhihua Dong Ye Li Zhiguo Wang Guangqiang Yin Xinzhong Wang Self-supervised recalibration network for person re-identification Defence Technology Person re-identification Attention mechanism Global information Local information Adaptive weighted fusion |
| title | Self-supervised recalibration network for person re-identification |
| title_full | Self-supervised recalibration network for person re-identification |
| title_fullStr | Self-supervised recalibration network for person re-identification |
| title_full_unstemmed | Self-supervised recalibration network for person re-identification |
| title_short | Self-supervised recalibration network for person re-identification |
| title_sort | self supervised recalibration network for person re identification |
| topic | Person re-identification Attention mechanism Global information Local information Adaptive weighted fusion |
| url | http://www.sciencedirect.com/science/article/pii/S2214914723000119 |
| work_keys_str_mv | AT shaoqihou selfsupervisedrecalibrationnetworkforpersonreidentification AT zhimingwang selfsupervisedrecalibrationnetworkforpersonreidentification AT zhihuadong selfsupervisedrecalibrationnetworkforpersonreidentification AT yeli selfsupervisedrecalibrationnetworkforpersonreidentification AT zhiguowang selfsupervisedrecalibrationnetworkforpersonreidentification AT guangqiangyin selfsupervisedrecalibrationnetworkforpersonreidentification AT xinzhongwang selfsupervisedrecalibrationnetworkforpersonreidentification |