[Project]SRGAN with WGAN, making super-resolution model training more stable

Hey guys,
Recently I open-source my project, using WGAN in SRGAN, SRGAN is an impressive super-resolution deep learning model and jeremy had introduced this paper in lesson14: Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network.

Since the SRGAN was first proposed in 2016, we haven’t had Wasserstein GAN(2017) yet, WGAN using wasserstein distance to measure the disturibution difference between different data set. As for the original GAN training, we don’t know when to stop training the discriminator or the generator, to get a nice result. But when using the wasserstein loss, as the loss decreasing, the result will be better. So what if we combine the WGAN and SRGAN?

My project is here: JustinhoCHN/SRGAN_Wasserstein, please check it out and starrrrrrrrrrr!
Thanks for reading.

initpintu6_%E5%89%AF%E6%9C%AC

1 Like

Great idea! Did you notice any improvements?

One advantage is that we don’t have to train the generator network and discriminator network separately, I think the model can converge faster, and it works really well in png image super-resolution.