It's not a issue, it's to help people reproduce it

[qianx77]

I found that although this is a single step diffusion, the memory consumption during training is still quite large. Considering that I only have RTX3090, I will record some of my settings, hoping to help those in need.

1、--train_batch_size set from 4 to 2.
2、also can set --gradient_checkpointing for less memory

the result can be reproduced !

[qianx77]

also need to set --gradient_checkpointing

[sadder971]

Hi, how did you prepare the test folder? When split="test" in PairedSROnlineTxtDataset(), it requires a folder for test_SR_bicubic and another folder for test_HR. How do we obtain the test_SR_bicubic one? Do we need to write our script to produce it?

[qianx77]

you just need to bulid these two folders and put your own data. by the way, the model is great!

[sadder971]

Are these data used during training, or are they for testing only?

[qianx77]

Are these data used during training, or are they for testing only?

up to you ， I only use it for testing