edoc

Empirically Analyzing the Effect of Dataset Biases on Deep Face Recognition Systems

Kortylewski, Adam and Egger, Bernhard and Schneider, Andreas and Gerig, Thomas and Morel-Forster, Andreas and Vetter, Thomas. (2018) Empirically Analyzing the Effect of Dataset Biases on Deep Face Recognition Systems. In: 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPR). pp. 2206-2215.

[img] PDF - Accepted Version
2054Kb

Official URL: https://edoc.unibas.ch/69078/

Downloads: Statistics Overview

Abstract

It is unknown what kind of biases modern in the wild face datasets have because of their lack of annotation. A di- rect consequence of this is that total recognition rates alone only provide limited insight about the generalization abil- ity of a Deep Convolutional Neural Networks (DCNNs). We propose to empirically study the effect of different types of dataset biases on the generalization ability of DCNNs. Us- ing synthetically generated face images, we study the face recognition rate as a function of interpretable parameters such as face pose and light. The proposed method allows valuable details about the generalization performance of different DCNN architectures to be observed and compared. In our experiments, we find that: 1) Indeed, dataset bias has a significant influence on the generalization performance of DCNNs. 2) DCNNs can generalize surprisingly well to un- seen illumination conditions and large sampling gaps in the pose variation. 3) Using the presented methodology we re- veal that the VGG-16 architecture outperforms the AlexNet architecture at face recognition tasks because it can much better generalize to unseen face poses, although it has sig- nificantly more parameters. 4) We uncover a main limita- tion of current DCNN architectures, which is the difficulty to generalize when different identities to not share the same pose variation. 5) We demonstrate that our findings on syn- thetic data also apply when learning from real-world data. Our face image generator is publicly available to enable the community to benchmark other DCNN architectures.
Faculties and Departments:05 Faculty of Science > Departement Mathematik und Informatik > Ehemalige Einheiten Mathematik & Informatik > Computergraphik Bilderkennung (Vetter)
UniBasel Contributors:Vetter, Thomas and Kortylewski, Adam and Gerig, Thomas and Morel, Andreas
Item Type:Conference or Workshop Item, refereed
Conference or workshop item Subtype:Conference Paper
Publisher:IEEE
Note:Publication type according to Uni Basel Research Database: Conference paper
Language:English
Related URLs:
Identification Number:
edoc DOI:
Last Modified:31 Jan 2020 13:54
Deposited On:27 Feb 2019 10:01

Repository Staff Only: item control page