FRoundation: Are foundation models ready for face recognition?

IF 4.2 3区 计算机科学 Q2 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE Image and Vision Computing Pub Date : 2025-02-19 DOI:10.1016/j.imavis.2025.105453
Tahar Chettaoui , Naser Damer , Fadi Boutros
{"title":"FRoundation: Are foundation models ready for face recognition?","authors":"Tahar Chettaoui ,&nbsp;Naser Damer ,&nbsp;Fadi Boutros","doi":"10.1016/j.imavis.2025.105453","DOIUrl":null,"url":null,"abstract":"<div><div>Foundation models are predominantly trained in an unsupervised or self-supervised manner on highly diverse and large-scale datasets, making them broadly applicable to various downstream tasks. In this work, we investigate for the first time whether such models are suitable for the specific domain of face recognition (FR). We further propose and demonstrate the adaptation of these models for FR across different levels of data availability, including synthetic data. Extensive experiments are conducted on multiple foundation models and datasets of varying scales for training and fine-tuning, with evaluation on a wide range of benchmarks. Our results indicate that, despite their versatility, pre-trained foundation models tend to underperform in FR in comparison with similar architectures trained specifically for this task. However, fine-tuning foundation models yields promising results, often surpassing models trained from scratch, particularly when training data is limited. For example, after fine-tuning only on 1K identities, DINOv2 ViT-S achieved average verification accuracy on LFW, CALFW, CPLFW, CFP-FP, and AgeDB30 benchmarks of 87.10%, compared to 64.70% achieved by the same model and without fine-tuning. While training the same model architecture, ViT-S, from scratch on 1k identities reached 69.96%. With access to larger-scale FR training datasets, these performances reach 96.03% and 95.59% for the DINOv2 and CLIP ViT-L models, respectively. In comparison to the ViT-based architectures trained from scratch for FR, fine-tuned same architectures of foundation models achieve similar performance while requiring lower training computational costs and not relying on the assumption of extensive data availability. We further demonstrated the use of synthetic face data, showing improved performances over both pre-trained foundation and ViT models. Additionally, we examine demographic biases, noting slightly higher biases in certain settings when using foundation models compared to models trained from scratch. We release our code and pre-trained models’ weights at <span><span>github.com/TaharChettaoui/FRoundation</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50374,"journal":{"name":"Image and Vision Computing","volume":"156 ","pages":"Article 105453"},"PeriodicalIF":4.2000,"publicationDate":"2025-02-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Image and Vision Computing","FirstCategoryId":"94","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S0262885625000411","RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q2","JCRName":"COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE","Score":null,"Total":0}
引用次数: 0

Abstract

Foundation models are predominantly trained in an unsupervised or self-supervised manner on highly diverse and large-scale datasets, making them broadly applicable to various downstream tasks. In this work, we investigate for the first time whether such models are suitable for the specific domain of face recognition (FR). We further propose and demonstrate the adaptation of these models for FR across different levels of data availability, including synthetic data. Extensive experiments are conducted on multiple foundation models and datasets of varying scales for training and fine-tuning, with evaluation on a wide range of benchmarks. Our results indicate that, despite their versatility, pre-trained foundation models tend to underperform in FR in comparison with similar architectures trained specifically for this task. However, fine-tuning foundation models yields promising results, often surpassing models trained from scratch, particularly when training data is limited. For example, after fine-tuning only on 1K identities, DINOv2 ViT-S achieved average verification accuracy on LFW, CALFW, CPLFW, CFP-FP, and AgeDB30 benchmarks of 87.10%, compared to 64.70% achieved by the same model and without fine-tuning. While training the same model architecture, ViT-S, from scratch on 1k identities reached 69.96%. With access to larger-scale FR training datasets, these performances reach 96.03% and 95.59% for the DINOv2 and CLIP ViT-L models, respectively. In comparison to the ViT-based architectures trained from scratch for FR, fine-tuned same architectures of foundation models achieve similar performance while requiring lower training computational costs and not relying on the assumption of extensive data availability. We further demonstrated the use of synthetic face data, showing improved performances over both pre-trained foundation and ViT models. Additionally, we examine demographic biases, noting slightly higher biases in certain settings when using foundation models compared to models trained from scratch. We release our code and pre-trained models’ weights at github.com/TaharChettaoui/FRoundation.

Abstract Image

查看原文
分享 分享
微信好友 朋友圈 QQ好友 复制链接
本刊更多论文
求助全文
约1分钟内获得全文 去求助
来源期刊
Image and Vision Computing
Image and Vision Computing 工程技术-工程:电子与电气
CiteScore
8.50
自引率
8.50%
发文量
143
审稿时长
7.8 months
期刊介绍: Image and Vision Computing has as a primary aim the provision of an effective medium of interchange for the results of high quality theoretical and applied research fundamental to all aspects of image interpretation and computer vision. The journal publishes work that proposes new image interpretation and computer vision methodology or addresses the application of such methods to real world scenes. It seeks to strengthen a deeper understanding in the discipline by encouraging the quantitative comparison and performance evaluation of the proposed methodology. The coverage includes: image interpretation, scene modelling, object recognition and tracking, shape analysis, monitoring and surveillance, active vision and robotic systems, SLAM, biologically-inspired computer vision, motion analysis, stereo vision, document image understanding, character and handwritten text recognition, face and gesture recognition, biometrics, vision-based human-computer interaction, human activity and behavior understanding, data fusion from multiple sensor inputs, image databases.
期刊最新文献
Editorial Board Early progression detection from MCI to AD using multi-view MRI for enhanced assisted living An edge-aware high-resolution framework for camouflaged object detection MUNet: A lightweight Mamba-based Under-Display Camera restoration network Adaptive scale matching for remote sensing object detection based on aerial images
×
引用
GB/T 7714-2015
复制
MLA
复制
APA
复制
导出至
BibTeX EndNote RefMan NoteFirst NoteExpress
×
×
提示
您的信息不完整,为了账户安全,请先补充。
现在去补充
×
提示
您因"违规操作"
具体请查看互助需知
我知道了
×
提示
现在去查看 取消
×
提示
确定
0
微信
客服QQ
Book学术公众号 扫码关注我们
反馈
×
意见反馈
请填写您的意见或建议
请填写您的手机或邮箱
已复制链接
已复制链接
快去分享给好友吧!
我知道了
×
扫码分享
扫码分享
Book学术官方微信
Book学术文献互助
Book学术文献互助群
群 号:481959085
Book学术
文献互助 智能选刊 最新文献 互助须知 联系我们:info@booksci.cn
Book学术提供免费学术资源搜索服务,方便国内外学者检索中英文文献。致力于提供最便捷和优质的服务体验。
Copyright © 2023 Book学术 All rights reserved.
ghs 京公网安备 11010802042870号 京ICP备2023020795号-1