Adaptive Hypermodels Usage in Person Re-Identification
DOI:
https://doi.org/10.31649/1997-9266-2025-179-2-138-146Keywords:
re-identification, hypernetworks, dynamic adaptation, incremental learning, CNN, Transformer, OSNet, Market-1501, DukeMTMC-ReID, MSMT17Abstract
This article deals with existing development and implementation approaches to the adaptive hypermodels usage for person re-identification. The research is based on a comparative study of the models performance applied to different data sets, covering both laboratory scenarios and real operating conditions. Main attention is paid to the assessment of key quality indicators, in particular, recognition accuracy (mAP, CMC) and information processing speed, which allows comprehensive coverage of the effectiveness of the methods used.
The study focuses on analyzing the impact of dynamic parameter changes on the results of model work, as well as on the study of incremental learning strategies that helps reduce the risk of catastrophic (knowledge loss) forgetting when adapting to new conditions without the need for complete re-training. Thanks to this approach, the system is able to quickly respond to changes in shooting conditions, for example, variations in lighting, angles and other characteristics, which is critically important for the video surveillance systems.
Based on the analysis done, promising areas of further research are outlined, aimed at improving adaptive learning algorithms, developing new architectural solutions and optimizing scaling processes. This, in its turn, will contribute to the implementation of more reliable and effective re-identification technologies in modern information systems.
The proposed approach combines usage of a hypermodel with an updated deep neural network, the key advantage of which is its high adaptability and stability of learning, ensured by the usage of a dynamic parameter adjustment, using hypermodels. Combination of cross-entropy and triplet losses allows us to effectively form compact and separate features for different identities, as well as increase the model ability to identify an object even in cases of significant variability of input data.
The results of the study demonstrate the prospects for integrating adaptive mechanisms into modern re-identification systems, providing increased resistance to changes in operating conditions and a high level of productivity, which is a necessary condition for successful practical application in information technologies.
References
О. М. Кириленко, «Розробка методу повторної ідентифікації людини,» Опт-ел. інф-енерг. техн., вип. 41, т. 1, с. 25-32, 2022.
D. Ha, A. Dai, and Q. V. Le, “Hyper Networks,” arXiv preprint. 2016. [Electronic resource]. Available: https://arxiv.org/abs/1609.09106 .
X. Jia, X. Wei, and X. Cao, “Dynamic Filter Networks,” arXiv preprint. 2016. [Electronic resource]. Available: https://arxiv.org/abs/1605.09673 .
Market-1501 dataset. [Electronic resource]. Available: https://www.pkuvmc.com/dataset.html .
DukeMTMC-ReID dataset [Electronic resource]. Available: https://github.com/layumi/DukeMTMC-reID_evaluation .
MSMT17 dataset. [Electronic resource]. Available: https://arxiv.org/abs/1711.08565 .
K. Zhou, Y. Yang, A. Cavallaro, and T. Xiang, “Learning Generalisable Omni-Scale Representations for Person Re-Identification,” IEEE Transactions on Pattern Analysis and Machine Intelligence, т. 44, no. 10, pp. 7593-7609, 2021.
A. Vaswani, et al., “Attention is All You Need,” Advances in Neural Information Processing Systems, pp. 5998-6008, 2017.
I. Goodfellow, Y. Bengio, and A. Courville, Deep Learning, Cambridge. MIT press, 2016, 775 c.
A. Hermans, L. Beyer, and B. Leibe, “In Defense of the Triplet Loss for Person Re-Identification,” arXiv preprint, 2017. [Electronic resource]. Available: https://arxiv.org/abs/1703.07737 .
E. Ristani, F. Solera, R. Zou, R. Cucchiara, and C. Tomasi, “Performance Measures and a Data Set for Multi-Target, Multi-Camera Tracking,” European Conference on Computer Vision (ECCV), pp. 17-35, 2016.
K. Zhou, and T. Xiang, Torchreid: A Library for Deep Learning Person Re-Identification in Pytorch.
L. Zheng, Y. Yang, and A. G. Hauptmann, “Person re-identification: Past, present and future,” arXiv preprint, 2016. [Electronic resource]. Available: https://arxiv.org/abs/1610.02984 .
J. Deng, W. Dong, R. Socher, L. Li, K. Li, and L. Fei-Fei, “ImageNet: A large-scale hierarchical image database,” IEEE Conference on Computer Vision and Pattern Recognition, 2009, pp. 248-255.
K. He, X. Zhang, S. Ren, and J. Sun Deep, “Residual Learning for Image Recognition,” IEEE Conference on Computer Vision and Pattern Recognition, 2016, pp. 770-778.
D. P. Kingma, and J. Ba, “Adam: A Method for Stochastic Optimization,” arXiv preprint, 2014. [Electronic resource]. Available: https://arxiv.org/abs/1412.6980 .
I. Loshchilov, and F. Hutter, “SGDR: Stochastic Gradient Descent with Warm Restarts,” arXiv preprint, 2016. [Electronic resource]. Available: https://arxiv.org/abs/1608.03983 .
Downloads
-
pdf (Українська)
Downloads: 2
Published
How to Cite
Issue
Section
License
This work is licensed under a Creative Commons Attribution 4.0 International License.
Authors who publish with this journal agree to the following terms:
- Authors retain copyright and grant the journal right of first publication.
- Authors are able to enter into separate, additional contractual arrangements for the non-exclusive distribution of the journal's published version of the work (e.g., post it to an institutional repository or publish it in a book), with an acknowledgment of its initial publication in this journal.
- Authors are permitted and encouraged to post their work online (e.g., in institutional repositories or on their website) prior to and during the submission process, as it can lead to productive exchanges, as well as earlier and greater citation of published work (See The Effect of Open Access).
