Why training my own data generates blurry images (with pictures)

[littlebig-balala]

I am using unet18 X4 pre-training model for training, but the generated results are poor, I checked my data without any problem.
I would like to ask what is the cause of this?
Below is a picture of the GT

and this is a picture of the FLAVR generate