/Understanding-Diffusion-Models

Primary LanguageJupyter Notebook

My learning path towards understanding Diffusion Models

1. Mnist DDPM & Improved DDPM

Epochs samples with posterior variance set to

${\sigma_t}^2 = \frac{1 - \bar{\alpha}{t-1}}{1 - \bar{\alpha{t}}}\beta_t$ ,   and   ${\sigma_t}^2 = \beta_t$

Alt Text Alt Text

2. Mnist DDIM with Diffusion AutoEncoder

Original samples (left) and restored samples (right)

Alt Text Alt Text