Training cyclegan from zero

I am trining the cyclegan from zero to see how it performs and i have noticed that after 5 epochs the “fake” images are purple or geen blury images. is it normal?
i am using a ghibili2photo dataset

Hi Alz!

I think it is quite normal to get this kind of output in your earlier epochs. But in my view, one cannot predict what’s “normal” with respect to a particular problem, dataset, or model architecture.
So you have to confirm if it is normal. How do you do that? I would suggest you: wait for a few more epochs and see if the results get better, and see if the model “converges” (this might give you an idea of what’s happening). If it improves then cool else you might have to address some internal issues.

Regards,
Nithin