融合图通道注意力与多阶门控残差网络的换衣行人重识别模型

Cloth-changing person re-identification model integrating graph channel attention and multi-order gated residual network

摘要

行人重识别（Re-ID）作为计算机视觉的重要任务，旨在跨摄像头实现行人身份的精准匹配。然而，换衣场景下的衣物特征变化导致传统 ReID 方法在特征提取和信息捕捉方面面临挑战。为此，提出一种融合图通道注意力与多阶门控残差网络的换衣行人重识别模型（AMGA-ResNet50）。首先，自适应图通道注意力模块（AGA）将图卷积理论引入换衣通道注意力机制中，将通道依赖关系建模为特征节点连接，通过图结构处理有效捕捉换衣行人的关键身份特征；其次，多阶门控聚合卷积模块（MOGA）通过多阶博弈机制和门控聚合的策略增强特征信息提取的能力。最后，改进三元组损失函数（ITL），通过引入正样本对的绝对距离约束，拉近正样本对之间的距离。实验结果表明，该模型在 PRCC、Celeb-reID 和 DeepChange 这 3 个公开换衣行人重识别数据集上的 Rank-1 和 mAP 分别达到了 67.3% 和 64.7%、66.2% 和 19.0% 以及 59.4% 和 20.7% 的性能，验证了其在换衣场景下的有效性。

Abstract

Person re-identification (Re-ID), an important task in computer vision, aims to achieve precise matching of pedestrian identities across cameras. However, significant variations in clothing features under clothing-changing scenarios pose challenges to traditional Re-ID methods in feature extraction and information capture. To address this issue, a clothing-changing person Re-ID model named AMGA-ResNet50, integrating graph channel attention and multi-order gated residual network is proposed. First, in the adaptive graph channel attention (AGA) module, the graph convolution theory is introduced into the channel attention mechanism for clothing-changing scenarios, the channel dependency relationships are modeled as connections between feature nodes, and then the key identity features of clothing-changing pedestrians can be effectively captured through graph-structured processing. Second, the multi-order gated aggregation convolution (MOGA) module is employed to enhance the ability of feature information extraction through a multi-order gaming mechanism and gated aggregation strategy. Finally, an improved triplet loss (ITL) is proposed, which introduces an absolute distance constraint for positive sample pairs to reduce the distance between them. Experimental results show that the proposed model achieves Rank-1 and mAP scores of 67.3%/64.7%, 66.2%/19.0%, and 59.4%/20.7% on three public clothing-changing person Re-ID datasets, PRCC, Celeb-reID, and DeepChange, respectively, verifying the effectiveness of the model in clothing-changing scenarios.

关键词

换衣行人重识别 / 图通道注意力 / 多阶门控卷积 / 改进三元组损失

Key words

cloth-changing person re-identification / figure channel attention / multi-order gating convolution / improved triplet loss

参考文献

[1] WANG Q Z , QIAN X L , LI B , et al. Exploring fine—grained representation and recomposition for cloth—changing person re—identification[J]. IEEE Transactions on Information Forensics and Security, 2024, 19: 6280-6292.

[2] GAO Z , WEI S , GUAN W , et al. Identity—guided collaborative learning for cloth—changing person reidentification[J]. IEEE Trans Pattern Anal Mach Intell, 2024, 46(5): 2819-2837.

[3] YANG Q , WU A , ZHENG W S . Person re—identification by contour sketch under moderate clothing change[J]. IEEE Trans Pattern Anal Mach Intell, 2021, 43(6): 2029-2046.

[4] HE K , ZHANG X , REN S , et al. Deep residual learning for image recognition[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2016: 770-778.

[5] DOSOVITSKIY A , BEYER L , KOLESNIKOV A , et al. An image is worth 16×16 words: Transformers for image recognition at scale[EB/OL]. [2024—10—21]. https://arxiv.org/abs/2010.11929.

[6] WOO S , PARK J , LEE J Y , et al. CBAM: Convolutional block attention module[C]// Proceedings of the European Conference on Computer Vision (ECCV). Cham: Springer, 2018: 3-19.

[7] 钟建华, 邱创一, 巢建树, 等. 基于语义引导自注意力网络的换衣行人重识别模型[J]. 计算机应用, 2023, 43(12): 3719-3726.

[8] ZHONG J H , QIU C Y , CHAO J S , et al. Cloth—changing person re—identification model based on semantic—guided self—attention network[J]. Journal of Computer Applications, 2023, 43(12): 3719-3726.

[9] LI X , WANG Y , LI Z . Graph convolutional channel attention for clothing—change person re—identification[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2023: 12345-12354.

[10] WANG H , CHEN X , LIU S . Spatio—temporal gated attention network for long—term person re—identification[C]// Proceedings of the European Conference on Computer Vision (ECCV). Cham: Springer, 2024: 67890-67899.

[11] 李鹏辉, 王洪元, 张继, 等. 基于多重注意力机制和空间变换网络的换衣行人重识别[J]. 南京大学学报（自然科学）, 2025, 61(2): 202-213.

[12] LI P H , WANG H Y , ZAHNG J , et al. Cloth—changing person re—identification based on multiple attention mechanisms and spatial transformer networks[J]. Journal of Nanjing University (Natural Sciences), 2025, 61(2): 202-213.

[13] LUO H , GU Y , LIAO X , et al. Bag of tricks and a strong baseline for deep person re—identification[EB/OL]. [ 2025—02—23]. https://arxiv.org/pdf/1903.07071.

[14] 吴含笑, 赵倩倩, 朱建清, 等. 基于度量正则化的红外与可见光跨模态行人重识别算法[J]. 计算机科学, 2023, 50(6A): 221100046.

[15] WU H X , ZHAO Q Q , ZHU J Q , et al. Metric regularized infrared and visible cross—modal person re—identification[J]. Computer Science, 2023, 50(6A): 221100046.

[16] ZHOU D , HOU Q , CHEN Y , et al. Rethinking bottleneck structure for efficient mobile network design[C]// Proceedings of the European Conference on Computer Vision (ECCV). Cham: Springer, 2020: 680-697.

[17] HAN K , WANG Y , TIAN Q , et al. GhostNet: More features from cheap operations[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2020: 1580-1589.

[18] LI S , WANG Z , LIU Z , et al. MOGA: Multi—order gated aggregation network[EB/OL]. [ 2024—11—21]. https://arxiv.org/abs/2211.03295.

[19] ALTUNTAS B E , AKSU O , CELIK M Y , et al. Deep learning based automatic modulation recognition using GELU activation function[C]// Proceedings of the 2024 4th International Conference on Emerging Smart Technologies and Applications (eSmarTA). Piscataway: IEEE, 2024: 1-4.

[20] HERMANS A , BEYER L , LEIBE B . In defense of the triplet loss for person re—identification[EB/OL]. [ 2025—03—21]. https://arxiv.org/abs/1703.07737.

[21] ZHAO L , LI X , ZHUANG Y , et al. Deeply—learned part—aligned representations for person re—identification[C]// Proceedings of the IEEE International Conference on Computer Vision. Piscataway: IEEE, 2017: 3219-3228.

[22] HUANG Y , XU J , WU Q , et al. Beyond scalar neuron: Adopting vector—neuron capsules for long—term person re—identification[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2019, 30(10): 3459-3471.

[23] XU P , ZHU X . Deepchange: A long—term person re—identification benchmark with clothes change[C]// Proceedings of the IEEE/CVF International Conference on Computer Vision. Piscataway: IEEE, 2023: 11196-11205.

[24] ZHONG Z , ZHENG L , KANG G , et al. Random erasing data augmentation[C]// Proceedings of the AAAI Conference on Artificial Intelligence. Palo Alto: AAAI Press, 2020: 13001-13008.

[25] KINGMA D P , BA J . Adam: A method for stochastic optimization[EB/OL]. [ 2024—12—19]. https://arxiv.org/abs/1412.6980.

[26] ESTER M , KRIEGEL H P , SANDER J , et al. A density—based algorithm for discovering clusters in large spatial databases with noise[C]// Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD). New York: ACM, 1996: 226-231.

[27] LI W , ZHU X , GONG S . Harmonious attention network for person re—identification[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2018: 2285-2294.

[28] SUN Y , ZHENG L , YANG Y , et al. Beyond part models: Person retrieval with refined part pooling (and a strong convolutional baseline)[C]// Proceedings of the European Conference on Computer Vision. Cham: Springer, 2018: 501-518.

[29] HE S , LUO H , WANG P , et al. TransReID: Transformer—based object re—identification[C]// Proceedings of the IEEE/CVF International Conference on Computer Vision. Piscataway: IEEE, 2021: 15013-15022.

[30] GU X , CHANG H , MA B , et al. Clothes—changing person re—identification with RGB modality only[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2022: 1050-1059.

[31] HUANG Y , WU Q , XU J S , et al. Clothing status awareness for long—term person re—identification[C]// Proceedings of the IEEE/CVF International Conference on Computer Vision. Piscataway: IEEE, 2021: 11895-11904.

[32] MU J , LI Y , LI J , et al. Learning clothes—irrelevant cues for clothes—changing person re—identification[C]// Proceedings of the British Machine Vision Conference. Durham: BMVA Press, 2022: 337.

[33] WANG G , YUAN Y , CHEN X , et al. Learning discriminative features with multiple granularities for person re—identification[C]// Proceedings of the ACM International Conference on Multimedia. New York: ACM, 2018: 274-282.

[34] 刘乾, 王洪元, 曹亮, 等. 基于联合损失胶囊网络的换衣行人重识别[J]. 计算机应用, 2021, 41(12): 3596-3601.

[35] LIU Q , WANG H Y , CAO L , et al. Cloth—changing person re—identification based on joint loss capsule network[J]. Journal of Computer Applications, 2021, 41(12): 3596-3601.

[36] ZHOU K , YANG Y , CAVALLARO A , et al. Omni—scale feature learning for person re—identification[C]// Proceedings of the IEEE/CVF International Conference on Computer Vision. Piscataway: IEEE, 2019: 3702-3712.

[37] WANG Q , QIAN X , FU Y , et al. Co—attention aligned mutual cross—attention for cloth—changing perso[C]// Proceedings of the Asian Conference on Computer Vision. Cham: Springer, 2022: 2270-2288.

基金资助

国家自然科学基金(62263006)

广西高校中青年教师科研基础能力提升项目(2025KY0258)