PFA-GAN: Pose Face Augmentation Based on Generative Adversarial Network

Bassel Zeno,Yuri Matveev,Ilya Kalinovskiy

doi:10.15388/21-infor443

Abstract

In this work, we propose a novel framework based on Generative Adversarial Networks for pose face augmentation (PFA-GAN). It enables a controlled pose synthesis of a new face image from a source face given a driving one while preserving the identity of the source face. We introduce a method for training the framework in a fully self-supervised mode using a large-scale dataset of unconstrained face images. Besides, some augmentation strategies are presented to expand the training set. The face verification experimental results demonstrate the effectiveness of the presented augmentation strategies as all augmented datasets outperform the baseline.

Highlights

A person’s face plays a key role in the identification of individual members of our highly social species due to delicate differences that make every human face unique
We present the Pose Face Augmentation GAN (PFA-GAN) that can transform a pose of a source face image using another face image while preserving the identity of the source image, as well as the pose and the expression of the driving face image
Inspired by the IP-GAN model, we present a novel framework (PFA-GAN) for pose face augmentation based on a generative adversarial network

Summary

Introduction

A person’s face plays a key role in the identification of individual members of our highly social species due to delicate differences that make every human face unique. Many remarkable works based on Deep Neural Networks have demonstrated unprecedented performance on several computer vision tasks, such as facial landmark detection, face identification, face verification, face alignment, emotion classification, etc They showed that achieving a good generalization in unconstrained conditions strongly relies on training them on large and complex datasets. Many learning-based methods have been proposed for face rotation, where most of them rely on a generative adversarial network (Tran et al, 2017; Tian et al, 2018; Cao et al, 2018a; Antoniou et al, 2018; Yin et al, 2017; Huang et al, 2017; Zeno et al, 2019a). Zeno et al (2019a) proposed IP-GAN framework to generate a face image of any specific identity with an arbitrary target pose by explicitly disentangling identity and pose representation from a face image

Objectives

Methods

Findings

Conclusion