search

LituRout / PSLD

Posterior Sampling using Latent Diffusion
124 stars 12 forks source link

Requirement for the set up for the Latent Diffusion model (LDM-VQ-4) #7

Closed csy2077 closed 4 months ago

csy2077 commented 4 months ago

Hi, I appreciate the nice job.

We also want to try the process of posterior sampling with the Latent Diffusion model (LDM-VQ-4) mentioned in your paper, but it seems that there is no code using the Latent Diffusion model (LDM-VQ-4).

Do you have a plan to release the code using the LDM-VQ-4?

LituRout commented 4 months ago

Hi csy2077,

Thanks for the comment. The code for posterior sampling using LDM-VQ-4 is already there in our codebase. Please take a look here: https://github.com/LituRout/PSLD/blob/main/stable-diffusion/run/inverse_sr_ldm.sh.

It is important to note that we have not carefully tuned the hyper-parameters for LDM-VQ-4 as our goal was to leverage SD-V1.5. The given hyper-parameters of LDM were suggested by another group of researchers. If these parameters do not work, you might want to tune the parameters depending on your application.

Best, Litu

csy2077 commented 4 months ago

Hi Litu,

Thanks a lot for your reply and your answer is really helpful !

Best, csy2077