Open ArmastusChen opened 1 year ago
@ArmastusChen Thank you for your attention! Achieving satisfactory results depends critically on the learning rate setting, which requires extensive experimental testing. Setting the learning rate is not only a matter of being too large or too small; it also needs to be adjusted at different time steps. Our experiments have shown that a good learning rate setting will be effective in generating most of the images. Currently, my suggestion is to only use guidance in the semantic stage, which may solve your problem to some extent! We will release our strategy for setting the learning rate later, and we hope you will follow our work.
Hi,
Thanks for the interesting work! I tried to re-implement your method using faceID as guidance. With the larger guidance weight, I got this wired output. But with the small weight, the result did not match the given condition.
Do you have similar observation? Any suggestion please? Thank you!