neilthefrobot
neilthefrobot
I am also wondering and found this - > Variable net can be squeeze, alex, vgg. Network alex is fastest, performs the best, and is the default. Set model=net for...
It does not use ground truth.
To successfully train SRGan you must start it out with an already trained SRResNet. Otherwise you will see your generator/discriminator losses very high and stuck in a local minima right...
All you need to change is get_gan_network() to be - `def get_gan_network(discriminator, shape, generator, optimizer): gan_input = Input(shape=shape) x = generator(gan_input) gan = Model(inputs=gan_input, outputs=x) gan.compile(loss='mse, optimizer=optimizer) return gan` And...
> Thanks for the code. So you are basically training a non-GAN this way? Once this part is trained, do you save and re-load the weights and then continue training...
The first loss returned from train_on_batch when there are multiple losses is always a sum of all the losses. The next number is the VGG perceptual loss (how close the...
The issue is definitely .jpg artifacts and I was able to get around this by taking my HR training set, down sampling 4x, then converting it to .jpg with a...
This is still an issue for me =( No matter what factor I use it does 2x
I spend hours trying to debug why loading a model doesn't work (I am on Windows) but the solution was simple. When you create a set() from data_ you have...