[WIP] Identity-conditioned Face Transformations Using Generative Adversarial Networks

Source: original report.

Introduction

This GAN receives two images

• input image A, and
• conditioning image B.

The output is a transformation of A in which some characteristics of the identity of B have been included.

TODO

• Implement main training loop
• Validation
• Better configuration
  • Configuration based on config file (--configuration <file>)
• Resume training (--continue <traning_config>)
• Scorer for tensorboard
• Long training
• Clean unused code from StarGAN

Technical report

This repository contains an implementation of the system proposed in Identity-conditioned Face Transformations Using Generative Adversarial Networks. The original work has been done during an internship in Fujitsu Laboratories LTD Japan as part of the Vulcanus in Japan programme and presented at Swopp 2018.

Nuevo Castro, G., Kasagi, A., Yamazaki, M., Tabaru, T. and Ike, A. (2018).
Identity-conditioned Face Transformations Using Generative Adversarial Networks.
In: SWoPP. [online] Available at: https://sites.google.com/site/swoppweb/.