do we literally need caption the reg image?
#1
by
Elaina617
- opened
i want to ask, do we need caption the reg image like the instance image?
In my case the regularization images are split into several categories. Therefore as the captioning part shows, to make the category trigger words effective the tags are probably necessary.
Otherwise I conjecture the tags "regularize the text encoder" and ensure the booru tags are mostly effective, but the difference probably only becomes observable after heavy training.