This repo contains some experiments with Denoising Diffusion Implicit Models. In particular, I play with simple distributions and see how DDIM translates latents to datapoints. This is a particularly interesting class of models because the latent space is intrinsic, it's not learned, and it's really cool to investigate.
u
ddim
public
28 stars
0 forks
0 issues
Commits
List of commits on branch master.Unverified
27950a639afbbe3de8f8f72d8605f14f91a5742aupdate notebook
uunixpickle committed 3 years ago
Unverified
148bc88995b7b55cbf8506423284226649591f54more conditioning experiments
uunixpickle committed 3 years ago
Unverified
a5442b7f3af11181b0291b98c55d941c158b7c13fix dtype issues
uunixpickle committed 3 years ago
Unverified
e67ea90a347dca9b90bf9e1778a4037ffbf3bfd3scale energy by sigma
uunixpickle committed 3 years ago
Unverified
64bdf87649715d2ed6d08ccd53bec2bc90b6002econditional sampling works in notebook
uunixpickle committed 3 years ago
Unverified
5a2e238351d8b0737ccaaa22b85ebf9cfb962147fix conditional sign, add temp
uunixpickle committed 3 years ago