about the train

[Jiaxin-lucky]

thanks for your wonderful work! in your paper, I find that you use the target y that is a degradation image as the GT, i am not understand why use this image as GT is enough for training the model to do restoration. actually, i think the model's output is similar to target y, not the original clean one. Hope for your answer!

[yohan-pg]

Hi Jiaxin, For evaluation purposes, we create targets from known ground truths images by degrading them, but in any practical use case the ground truths are unknown. There is no training process here, everything is unsupervised.