Ours (w/fine-tune all) ？

[wsfk716]

I just want to ask, since the Ours (w/fine-tune all) method mentioned in the paper is to fine-tune all parameters, what is the difference between it and Dreambooth? I don't quite understand it.

[nupurkmr9]

Hi, in Ours (w/ fine-tune all) we also fine-tune the V* token, use real images as regularization, and use the data augmentation technique as described in the paper.
Compared to this in DreamBooth, we used generated images as regularization, the V* token is not optimized, and no data augmentation is used.
Hope this is clear now.
Thanks!

[wsfk716]

合作

Ok OK, thank you very much for your answer!