deepak112 / Keras-SRGAN

Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network implemented in Keras
276 stars 146 forks source link

gan-loss #32

Open liangzifei opened 4 years ago

liangzifei commented 4 years ago

Dear, I saw three values in printing gan-loss: [0.020138385, 0.018311221, 1.8271642]

Could you please tell me what loss they are represented ?

Best

BassantTolba1234 commented 3 years ago

please , did you get the answer ?

liangzifei commented 3 years ago

The first is the sum loss A+0.001*B, second is generate loss, third is discriminator loss.

BassantTolba1234 commented 3 years ago

Please all, I need the code of implementation this part the part is

{The SRResNet networks were trained with a learning rate of 10−4 and 106 update iterations. We employed the trained MSE-based SRResNet network as initialization for the generator when training the actual GAN to avoid undesired local optima.{

neilthefrobot commented 3 years ago

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 output is to the target output in VGG space) The last number is the GAN loss (how well the generated images are fooling the discriminator)