Constructing Adaptive Multi-Scale Feature via Transformer-Aware Patch for Occluded Person Re-Identification
Person re-identification (Re-ID) aims to retrieve a specific pedestrian across a multi-disjoint camera in a surveillance system. Most of the research is based on a strong assumption that images should contain a full human torso. However, it cannot be guaranteed that all the people have a clear foreg...
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2022-07-01
|
| Series: | Symmetry |
| Subjects: | |
| Online Access: | https://www.mdpi.com/2073-8994/14/7/1454 |
| _version_ | 1797415376060940288 |
|---|---|
| author | Zhi Liu Xingyu Mu Shidu Dong Yunhua Lu Mingzi Jiang |
| author_facet | Zhi Liu Xingyu Mu Shidu Dong Yunhua Lu Mingzi Jiang |
| author_sort | Zhi Liu |
| collection | DOAJ |
| description | Person re-identification (Re-ID) aims to retrieve a specific pedestrian across a multi-disjoint camera in a surveillance system. Most of the research is based on a strong assumption that images should contain a full human torso. However, it cannot be guaranteed that all the people have a clear foreground because they are out of constraint. In the real world, a variety of occluded situations frequently appear in video monitoring, which impedes the recognition process. To settle the occluded person Re-ID issue, a new Dual-Transformer symmetric architecture is proposed in this work, which can reduce the occluded impact and build a multi-scale feature. There are two contributions to our proposed model. (i) A Transformer-Aware Patch Searching (TAPS) module is devised to learn visible human region distribution using a multiheaded self-attention mechanism and construct a branch of distributed information attention scale. (ii) An Adaptive Visible-Part Cropping (AVPC) Strategy, with two steps of cropping and weakly-supervised learning, is used to generate a fine-scale visible image for another branch. Only ID labels are utilized to restrain TAPS and AVPC without any extra visible-part annotation. Extensive experiments are conducted on two occluded person Re-ID benchmarks, confirming that our approach performs a SOTA or comparable effect. |
| first_indexed | 2024-03-09T05:47:45Z |
| format | Article |
| id | doaj.art-fcf380ba7888438d907080833b893618 |
| institution | Directory Open Access Journal |
| issn | 2073-8994 |
| language | English |
| last_indexed | 2024-03-09T05:47:45Z |
| publishDate | 2022-07-01 |
| publisher | MDPI AG |
| record_format | Article |
| series | Symmetry |
| spelling | doaj.art-fcf380ba7888438d907080833b8936182023-12-03T12:20:10ZengMDPI AGSymmetry2073-89942022-07-01147145410.3390/sym14071454Constructing Adaptive Multi-Scale Feature via Transformer-Aware Patch for Occluded Person Re-IdentificationZhi Liu0Xingyu Mu1Shidu Dong2Yunhua Lu3Mingzi Jiang4School of Artificial Intelligence, Chongqing University of Technology, Chongqing 401120, ChinaSchool of Artificial Intelligence, Chongqing University of Technology, Chongqing 401120, ChinaSchool of Artificial Intelligence, Chongqing University of Technology, Chongqing 401120, ChinaSchool of Artificial Intelligence, Chongqing University of Technology, Chongqing 401120, ChinaSchool of Artificial Intelligence, Chongqing University of Technology, Chongqing 401120, ChinaPerson re-identification (Re-ID) aims to retrieve a specific pedestrian across a multi-disjoint camera in a surveillance system. Most of the research is based on a strong assumption that images should contain a full human torso. However, it cannot be guaranteed that all the people have a clear foreground because they are out of constraint. In the real world, a variety of occluded situations frequently appear in video monitoring, which impedes the recognition process. To settle the occluded person Re-ID issue, a new Dual-Transformer symmetric architecture is proposed in this work, which can reduce the occluded impact and build a multi-scale feature. There are two contributions to our proposed model. (i) A Transformer-Aware Patch Searching (TAPS) module is devised to learn visible human region distribution using a multiheaded self-attention mechanism and construct a branch of distributed information attention scale. (ii) An Adaptive Visible-Part Cropping (AVPC) Strategy, with two steps of cropping and weakly-supervised learning, is used to generate a fine-scale visible image for another branch. Only ID labels are utilized to restrain TAPS and AVPC without any extra visible-part annotation. Extensive experiments are conducted on two occluded person Re-ID benchmarks, confirming that our approach performs a SOTA or comparable effect.https://www.mdpi.com/2073-8994/14/7/1454person re-identificationmultiheaded self-attentionweakly-supervised learningmulti-scale feature |
| spellingShingle | Zhi Liu Xingyu Mu Shidu Dong Yunhua Lu Mingzi Jiang Constructing Adaptive Multi-Scale Feature via Transformer-Aware Patch for Occluded Person Re-Identification Symmetry person re-identification multiheaded self-attention weakly-supervised learning multi-scale feature |
| title | Constructing Adaptive Multi-Scale Feature via Transformer-Aware Patch for Occluded Person Re-Identification |
| title_full | Constructing Adaptive Multi-Scale Feature via Transformer-Aware Patch for Occluded Person Re-Identification |
| title_fullStr | Constructing Adaptive Multi-Scale Feature via Transformer-Aware Patch for Occluded Person Re-Identification |
| title_full_unstemmed | Constructing Adaptive Multi-Scale Feature via Transformer-Aware Patch for Occluded Person Re-Identification |
| title_short | Constructing Adaptive Multi-Scale Feature via Transformer-Aware Patch for Occluded Person Re-Identification |
| title_sort | constructing adaptive multi scale feature via transformer aware patch for occluded person re identification |
| topic | person re-identification multiheaded self-attention weakly-supervised learning multi-scale feature |
| url | https://www.mdpi.com/2073-8994/14/7/1454 |
| work_keys_str_mv | AT zhiliu constructingadaptivemultiscalefeatureviatransformerawarepatchforoccludedpersonreidentification AT xingyumu constructingadaptivemultiscalefeatureviatransformerawarepatchforoccludedpersonreidentification AT shidudong constructingadaptivemultiscalefeatureviatransformerawarepatchforoccludedpersonreidentification AT yunhualu constructingadaptivemultiscalefeatureviatransformerawarepatchforoccludedpersonreidentification AT mingzijiang constructingadaptivemultiscalefeatureviatransformerawarepatchforoccludedpersonreidentification |