Skoči na glavni sadržaj

Izvorni znanstveni članak

https://doi.org/10.24138/jcomss-2021-0035

The Effect of Latent Space Dimension on the Quality of Synthesized Human Face Images

Ivana Marin orcid id orcid.org/0000-0003-4869-6724 ; Faculty of Science, University of Split, Croatia
Sven Gotovac ; Faculty of Electrical Engineering, Mechanical Engineering and Naval Architecture, University of Split, Croatia
Mladen Russo ; Faculty of Electrical Engineering, Mechanical Engineering and Naval Architecture, University of Split, Croatia
Dunja Božić-Štulić ; Faculty of Electrical Engineering, Mechanical Engineering and Naval Architecture, University of Split, Croatia


Puni tekst: engleski pdf 10.537 Kb

str. 124-133

preuzimanja: 2.110

citiraj


Sažetak

In recent years Generative Adversarial Networks (GANs) have achieved remarkable results in the task of realistic image synthesis. Despite their continued success and advances, there still lacks a thorough understanding of how precisely GANs map random latent vectors to realistic-looking images and how the priors set on the latent space affect the learned mapping. In this work, we analyze the effect of the chosen latent dimension on the final quality of synthesized images of human faces and learned data representations. We show that GANs can generate images plausibly even with latent dimensions significantly smaller than the standard dimensions like 100 or 512. Although one might expect that larger latent dimensions encourage the generation of more diverse and enhanced quality images, we show that an increase of latent dimension after some point does not lead to visible improvements in perceptual image quality nor in quantitative estimates of its generalization abilities.

Ključne riječi

Generative Adversarial Networks, Latent space exploration, Latent dimension, Evaluation, Frechet Inception Distance (FID), Image synthesis

Hrčak ID:

259385

URI

https://hrcak.srce.hr/259385

Datum izdavanja:

30.6.2021.

Posjeta: 2.771 *