Closed wangskyone closed 1 year ago
Hi @wangskyone, I think what you are refering is indeed the loss, which equals to the squared point wise difference of the output y and the sample x which are multiplied point wise for the mask as per D1 setting. The if and else statement of your paper screenshoot are the if and elif statements of the code.
I mean according to the code x0 in the paper corresponds to $x_{std}$.
eposilon_theta
-> y
z -> $x_{std}$.
you are right, in terms of named variables, noise input in code with ground truth in paper overlaps, altought we didn't tried to match names when the pseudocode was written. Nevertheless, It will be easier to update the repo than the paper, so, as soon as possible I will update the variables names. Thanks for it.
Thanks for replying patiently.
In the paper, the model output is written as loss with ground truth