Data Imbalance in Large-Scale Face Recognition B y k l ekli Y z Tanimada Veri Dengesizligi

Date

2025-01-01

Author

Erturkmen, Ahmet
Akkaya, Kursat
Boncu, Ece Selin
Ozden, Harun
Akbas, Emre
Alatan, Abdullah Aydın

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

758
views

0
downloads

Cite This

In this study, proposals have been presented to improve various recognition problems encountered in large-scale face databases related to the open-set face recognition problem. Based on the experiments conducted, the most successful face detection model was determined to be SCRFD-BNKPS. For the face recognition problem, a embedding vector extraction model that provides high discrimination capability was selected from among the ArcFace models. The biggest issue observed when these extracted feature vectors were classified into identity classes using popular nonlinear classifiers was the varying number of identities within each class. To address this, the ROS and ENS algorithms were employed to ensure that classes with a small number of samples provided sufficient feedback to the classifier. Additionally, the k-NN algorithm, used to avoid the need for training a wholesale classifier in the open-set classification problem, was improved using the LMNN metric learning method.

Subject Keywords

data imbalance, face detection, face recognition, open set

URI

https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=105015577944&origin=inward
https://hdl.handle.net/11511/115972

Collections

Department of Electrical and Electronics Engineering, Conference / Seminar