Efficient Face Recognition Using Embedding-Based Distillation
DOI:
https://doi.org/10.53907/enpesj.v5i2.335Keywords:
Face Recognition, Knowledge Distillation, Mobile devices, Efficient Deep learningAbstract
Knowledge distillation (KD) facilitates the compression of large, high-performing neural networks into efficient
student models, enabling deployment on resource-limited devices like mobile phones and IoT systems. This paper
introduces a KD methodology, which involves training a student to capture a teacher’s soft labels, intermediate feature
representations, and ground truth labels, ensuring both compactness and accuracy. Applied to face recognition, our hybrid
KD framework trains a MobileFaceNet student under an InceptionResNetV1 teacher, achieving 90.40% accuracy and a
96.25% AUC, outperforming lightweight models while remaining suitable for edge devices. These results highlight the
potential of KD to enable robust, scalable face recognition solutions for real-world, resource-constrained environments.
Downloads
Published
How to Cite
Issue
Section
License
Copyright (c) 2025 ENP Engineering Science Journal
This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.
Articles published under a Publisher Office user license are protected by copyright. Users may access, download, copy, translate, text and data mine the articles for non-commercial purposes provided that users:
- Cite the article using an appropriate bibliographic citation (i.e. author(s), journal, article title, volume, issue, page numbers, DOI and the link to the definitive published version)
- Maintain the integrity of the article
- Retain copyright notices and links to these terms and conditions so it is clear to other users what can and cannot be done with the article
- Ensure that, for any content in the article that is identified as belonging to a third party, any re-use complies with the copyright policies of that third party
- Any translations, for which a prior translation agreement with Publisher Office has not been established, must prominently display the statement: "This is an unofficial translation of an article that appeared in a Publisher Office publication. Publisher Office has not endorsed this translation."
This is a non commercial license where the use of published articles for commercial purposes is prohibited. Commercial purposes include:
- Copying or downloading articles, or linking to such postings, for further redistribution, sale or licensing, for a fee
- Copying, downloading or posting by a site or service that incorporates advertising with such content
- The inclusion or incorporation of article content in other works or services (other than normal quotations with an appropriate citation) that is then available for sale or licensing, for a fee
- Use of articles or article content (other than normal quotations with appropriate citation) by for-profit organizations for promotional purposes, whether for a fee or otherwise.
- Use for the purposes of monetary reward by means of sale, resale, license, loan, transfer or other form of commercial exploitation.
