Open FromA2Z opened 1 year ago
@FromA2Z Thank you for your attention to our work! Regarding rho
or the learning rate, it is an important parameter used to control the step size of guidance. Poorly chosen strategies for its setting can result in undesirable generation outcomes. The two strategies you've identified are two of the relatively better ones we've found:
rho
based on the length of guidance or score.at.sqrt()
is a promising setting. However, it requires stopping the guidance after a certain number of steps, which is what the stop
parameter accomplishes.A detailed description of parameter settings is provided in the camera-ready version. We suggest you experiment with different parameter settings using the code we've provided. This should lead to numerous new discoveries. Thank you~
Hello, I am interested in the code you posted, thank you for sharing. What puzzles me is that there is not much discussion of scale factor in the paper. In SD Style, rho appears to be a learning rate associated with both "grad" and "classification guided effects",as shown below However, in Face ID, rho is equal to at.sqrt(), as follows: So, how exactly do we set up RHO, and is there some mathematical theory to support it? Thank you?