Individual Module Project, University of Potsdam
The purpose of this project was to apply cycle-consistent adversatial networks proposed by Zhu et al. [1] to the face attribute manipulation problem.
We investigate the use of unpaired image-to-image translation using CycleGAN to the task of eyeglasses removal from faces along with the reverse task of adding eyeglasses to facial images.
Final version can be found in keras folder. All implementation details and model architecture can also be found in the project paper (report_cycleGAN.pdf).
Data folders are not being uploaded to GitHub due to size issues and a little bit of privacy.
Sets:
-
Eyeglasses: 1777 training images
-
No-eyeglasses: 1687 training images
[1] Zhu, J. Y., Park, T., Isola, P., & Efros, A. A. (2017). Unpaired image-to-image translation using cycle-consistent adversarial networks.