Improved Pedestrian Attribute Recognition Algorithm Based on Image Style Transfer

DOI: https://doi.org/10.62517/jbdc.202501215

Author(s)

Senlin Zhang1, Lingyu Zhao1, Wenkai Ren1, Wanwan Wang2,*, Jiangang Zhang2

Affiliation(s)

1College of Artificial Intelligence and Big Data, Henan University of Technology, Zhengzhou, Henan, China 2Anhui USTC iFLYTEK Co., Ltd., Anhui, Hefei, China *Corresponding Author

Abstract

As an important research direction in the field of computer vision, pedestrian attribute recognition has a wide range of value in video surveillance and other applications. However, the existing models have the problem of insufficient generalization ability when testing in new scenes. In this paper, we propose an improved method for pedestrian attribute recognition based on CycleGAN image style transfer. The source domain training data (PA100K) is transformed into the target scene (RAP) style in an unsupervised manner, and an enhanced dataset with both target domain visual features and source domain annotation information is constructed. The experimental results show that the model fused with style transfer data significantly improves the accuracy of attribute recognition on RAP test set, in which gender recognition is improved by 3%, and shirt color recognition is improved by 12%, which verifies the effectiveness of the method in cross-domain adaptation. This study not only avoids the high cost of target scene data labeling, but also provides an effective solution for the scene transfer of pedestrian attribute recognition models.

Keywords

Pedestrian Attribute; Image Style Transfer; CycleGAN; ResNet; Data Augmentation

References

[1]Cao Yuran, Lu Weiqing, Yu Jinzuo, et al. Review and Prospect of Pedestrian Attribute Recognition Methods Based on Single Frame and Video Data in Surveillance Scenarios. Journal of Computer-Aided Design & Computer Graphics, 2024, 36(03): 336-356. [2]Zheng L, Zhang H, Sun S, et al. Person re-identification in thewild//Proceedings of the IEEE conference on computer vision andpattern recognition,2017:1367-1376. [3]Liu X, Zhao H, Tian M, et al. Hydraplus-net: Attentive deep features forpedestrian analysis//Proceedings of the IEEE international conferenceon computer vision,2017:350-359. [4]Wang J, Zhu X, Gong S, et al. Attribute Recognition by Joint Recurrent Learning of Context and Correlation//2017 IEEE International Conference on Computer Vision (ICCV). IEEE, 2017: 531-540. [5]Jian Jia, Xiao-Tang Chen & Kai-qi Huang. (2022). Pedestrian attribute recognition in surveillance scenes: a survey. Acta Computerica Sinica,45(08),1765-1793. [6]Xu C, Li W, Chen Z B. Scarcity-GAN: Scarce data augmentation for defect detection via generative adversarial nets. Neurocomputing, 2024, 566 (Jan. 21): 127061.1 to 127061.12. [7]Sharma, G., & Jurie, F. (2011). Learning discriminative spatial representation for image classification. Procedings of the British Machine Vision Conference 2011, 6.1-6.11. [8]CHEN H, GALLAGHER A, GIROD B. Describing Clothing by Semantic Attributes//Computer Vision – ECCV 2012, Lecture Notes in Computer Science. 2012: 609-623. [9]Zhu J, Liao S, Yi D, et al. Multi-label CNN based pedestrian attribute learning for soft biometrics//International Conference on Biometrics. IEEE, 2015:535-540. [10]Deng Y, Luo P, Loy C C, et al. Pedestrian Attribute Recognition at Far Distance. ACM, 2014, 22(ACM MM '14):789-792. [11]Li D, Zhang Z, Chen X, et al. A Richly Annotated Dataset for Pedestrian Attribute Recognition. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2016, 39(10): 2033-2047. [12]Li Y, Huang C, Loy C C, et al. Human Attribute Recognition by Deep Hierarchical Contexts. Lecture Notes in Computer Science, 2016, 9907: 480-495. [13]Liu X, Zhao H, Tian M, et al. HydraPlus-Net: Attentive Deep Features for Pedestrian Analysis//2017 IEEE International Conference on Computer Vision (ICCV). IEEE, 2017: 5078-5087. [14]He K, Zhang X, Ren S, et al. Deep Residual Learning for Image Recognition. IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016: 770-778.