Skip to main content
tetamusha
Projects
wgan_celeba
Reports
GAN training on image datasets

GAN training on image datasets

Douglas De Rizzo Meneghetti
Created on February 4|Last edited on February 4
This page displays the results of me learning to build and train GANs on various image datasets.
The code is available at https://github.com/douglasrizzo/mygan

WGAN on the CelebA dataset

The CelebA dataset was acquired here.
The images below are samples from the generator, generated from the same set of 25 noise vectors. We can gradually see an improvement in face generation.

Samples
card

We can also see the generator loss (top boundary) going down and the mean critic loss (top boundary) going up, which, according to my studies, indicates convergence of the model.
Critic loss is a mean of 5 critic training cycles for each generator training cycle, which makes it less noisy.



Created with ❤️ on Weights & Biases.

https://wandb.ai/tetamusha/wgan_celeba/reports/GAN-training-on-image-datasets--Vmlldzo2NzMxODkx
Made with Weights & Biases. Sign up or log in to create reports like this one.