Artificial intelligence and machine learning
Research Article
Using domain adaptation for the human pose estimation task
Andrei Sergeevich Tokarev1
, Ilia Mikhailovich Voronkov2
| 1,2 | Moscow Institute of Physics and Technology, Moscow, Russia |
| 1 |
tokarev.as@phystech.su |
Abstract. The article studies domain adaptation algorithms for the task of recognizing key points on the human body for the purpose of using them in sports, when it is necessary to increase the accuracy of recognition and reduce the labor intensity of manual data labeling. The result of the work is an algorithm for iterative adaptation of the model on its own pseudo-labels. It is experimentally shown that the method allows obtaining a more effective final neural network model in comparison with conventional additional training. (Linked article texts in English and in Russian).
Keywords: Keypoints, human pose estimation, unsupervised domain adaptation
For citation: Andrei S. Tokarev, Ilia M. Voronkov. Using domain adaptation for the human pose estimation task. Program Systems: Theory and Applications, 2025, 16:4, pp. 23–50. (in Engl. In Russ.). https://psta.psiras.ru/2025/4_23-50.
Full text of bilingual article (PDF): https://psta.psiras.ru/read/psta2025_4_23-50.pdf (Clicking on the flag in the header switches the page language).
The article was submitted 20.12.2024; approved after reviewing 16.01.2025; accepted for publication 07.04.2025; published online 30.08.2025.
