search

zelaki / DreamSound

Code for Investigating Personalization Methods in Text to Music Generation
https://zelaki.github.io/
28 stars 5 forks source link

prior preservation loss in Dreambooth #4

Open fundwotsai2001 opened 8 months ago

fundwotsai2001 commented 8 months ago

Just want to know that, are the result in the paper using prior preservation loss in Dreambooth? I saw your code have prior preservation loss, but when I run it in dreambooth_audioldm.py, it works, but in dreambooth_audioldm2.py there are some bugs.

fundwotsai2001 commented 8 months ago

image

zelaki commented 8 months ago

Hi, The results we include in the paper are obtained using AudioLDM with prior preservation loss. Thank you for noticing the bug in the code for AudioLDM2 I will take a look at it. Thodoris

fundwotsai2001 commented 8 months ago

There are also bugs when using classifier guidance.