| Issue |
ITM Web Conf.
Volume 78, 2025
International Conference on Computer Science and Electronic Information Technology (CSEIT 2025)
|
|
|---|---|---|
| Article Number | 02029 | |
| Number of page(s) | 7 | |
| Section | Machine Learning Applications in Vision, Security, and Healthcare | |
| DOI | https://doi.org/10.1051/itmconf/20257802029 | |
| Published online | 08 September 2025 | |
Analysis Loss Functions for Recognition Accuracy Improvement in Face Recognition
Faculty of Data Science, City University of Macau, Macau, 999078, China
In the field of face recognition, improving recognition accuracy is an important research direction. Due to the fact that face images may be disturbed during acquisition, transmission and storage, the recognition algorithms are not robust enough to meet the requirements of practical applications. Therefore, the introduction of an effective loss function is particularly critical. In this paper, the advantages and disadvantages of different loss functions will be derived by analysing the softmax, ArcFace and AdaFace loss functions and comparing the differences between them. Through the analysis, this paper concludes that the loss functions still have different defects in the application of real scenarios and cannot effectively identify data with fewer samples. For example, Arcace loss is not able to perfectly make the samples evenly distributed on the hypersphere in real life. Therefore, in the future design and optimisation of the self-supervised pre-training framework should be strengthened to make a greater improvement in the recognition accuracy of face recognition.
© The Authors, published by EDP Sciences, 2025
This is an Open Access article distributed under the terms of the Creative Commons Attribution License 4.0, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.
Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.
Initial download of the metrics may take a while.