Key-Locked-Rank-One-Editing-for-Text-to-Image-Personalization
Key-Locked-Rank-One-Editing-for-Text-to-Image-Personalization copied to clipboard
Questions about the reason for applying EMA to the target input i*
Firstly, I would like to extend my congratulations on your acceptance into SIGGRAPH. I find your work quite interesting and appreciate your contribution to the field.
And I got a question in Section 4.4 in your paper, I am curious to know the rationale behind using the Exponential Moving Average (EMA) for the target input (i.e., $i_* = 0.99i_* + 0.01 e_\text{concept}$), instead of directly using $i_* = e_\text{concept}$. In EMA, the $e_\text{concept}$ just accounts for such a small part. I am keen to understand the intuition underlying the application of EMA in this context.
Thank you for your time and consideration. I look forward to hearing back from you.