questions about OOE training

[SolitaryManF]

Can the training process of the whole model be divided into two steps：
(1)firstly, train the generator which uses FxSR model ;
(2)then, train Cψ with generator parameters fixed;

In your paper, section3.3 OOE, you said "Specifically, we generate a set of 21 SR results by changing t from 0 to 1 with a step of 0.05, and the optimal objective maps are generated by selecting the t with the lowest LPIPS among them for each pixel".
(1)Is 21 SR results generated by the generator FxSR（which can change t from 0 to 1）？
(2)about GT of predictive model Cψ：(T∗S), how did you obtain it? by mix pixel(which has the lowest LPIPS) of 21 SR results?

Hope to get a reply，thank you!

[seungho-snu]

Sorry for late reply.
Yes, As you mentioned, the training process of the whole model be divided into two steps.

(1)Is 21 SR results generated by the generator FxSR（which can change t from 0 to 1）？
-> Yes, t={0, 0.05,...0.95,1.0}
(2)about GT of predictive model Cψ：(T∗S), how did you obtain it? by mix pixel(which has the lowest LPIPS) of 21 SR results?
-> Yes, as you said, by mixing pixels of 21 SR results?

Thank you for you interest.