MASTER Assignment
producing high quality face morphs using gans
Type : Master M-CS
Period: Jan, 2019- Aug, 2020
Student : Meijer, G. (Gerben, Student M-CS)
Date Final project: August 18, 2020
Supervisors:
Abstract:
This work aims to improve the face morphing performance of MorGAN, the current state of the art in GAN-based face morphing. It introduces four possible improvements and evaluates their effect on morphing performance as well as image quality. Two of these possible improvements are shown to improve morphing performance. The first improvement introduces a feature-wise reconstruction loss to replace the pixel-wise reconstruction loss used by MorGAN. The results show that this feature-wise reconstruction loss can be used to achieve significantly better morphing performance with a smaller impact on image quality compared to the pixel-wise loss. The second improvement optimizes the morphs generated by already trained models by applying gradient descent on the latent vector of the morph. This technique improves morphing performance with minimal image quality loss but does increase the computational costs of morphing significantly.