Face Aging with Modified Identity-Preserved Conditional GANs

1. Introduction

The project aims to implement Identity-Preserved Conditional Generative Adversarial Networks (IPCGANs) to create synthetic images of people that will show how people will look after a certain number of years are passed. Previously, a similar problem has been solved using Cycle Gan, in which an image of a person has been used as input, and both aged and de-aged (depending on the condition) synthetic photo of a person has been returned. Now, we will make the project more complicated and powerful by training the model to output not only an aged and de-aged version of a person but several images corresponding to different age groups of the same person. For example, we will input an image of a 45-year-old person and generate 5 synthetic images corresponding to 5 age groups, the 20s, 30s,40s, 50, and 60s, thus simulating how one’s face will change the years.