Open Chenrf1121 opened 1 month ago
Thanks for your interest in our DSG. Our implementation is based on DDIM with eta=1, which is equal to DDPM. If you have to use the code in DDPM, I recommend returning posterior_mean and variance separately (like class DDPM(Line 403-420) in Linear_Inverse_Problems/guided_diffusion/gaussian_diffusion.py), then apply DSG (like class DSG(Line 101-136) in Linear_Inverse_Problems/guided_diffusion/conditional_methods.py).
Can we apply DSG on DDPM? like add some code in
in
Non_linear_problems/SD_style/ldm/models/diffusion/ddpm.py
line243-250. If yes, would you provide an example?