search

titu1994 / Super-Resolution-using-Generative-Adversarial-Networks

An implementation of SRGAN model in Keras
283 stars 85 forks source link

A Simple Confusion Regarding the Description Provided for Training the Full Model #27

Closed rmalav15 closed 7 years ago

rmalav15 commented 7 years ago

Hey @titu1994

Much much thanks for this great implementation.

I am slightly confused with following two lines.

To train the full network (Does NOT work properly right now, Discriminator is not correctly trained):

(Last lines in USAGE) and

Properly train SRGAN (SR ResNet + VGG + Discriminator) model.

(Right marked in PLANS)

Can you please let me know the current situation of full training?
Is there any implementation fault for full training or just it doesn't work properly after following everything correctly?

Thanks.

titu1994 commented 7 years ago

I've stopped working on the code, but i think the Discriminator now trains properly. I forgot to update the comment in the usage.

Even so, it's performance is abysmal. I suggest taking a look at neural enhance. It is the same model, but has pretrained weights and working scripts and docker installs.