search

eriklindernoren / PyTorch-GAN

PyTorch implementations of Generative Adversarial Networks.
MIT License
16.17k stars 4.05k forks source link

SRGAN - Adversarial Loss function #176

Open Sumeet2807 opened 2 years ago

Sumeet2807 commented 2 years ago

I see the loss implemented here are as follows : -

Losses

criterion_GAN = torch.nn.MSELoss() criterion_content = torch.nn.L1Loss()

It is not same as the original paper, right ? Adversarial loss should be a cross entropy loss if I am not wrong. Does this setup perform better than the original paper formulation ?