A gentle introduction to diffusion models without the math derivations, but rather, a focus on the concepts that define the diffusion models as described in the DDPM paper.
Full code and PDF slides available at: https://github.com/hkproj/pytorch-ddpm
Chapters
00:00 - Introduction
00:46 - Generative models
03:51 - Latent space
07:35 - Forward and reverse process
09:00 - Mathematical definitions
13:00 - Training loop
15:05 - Sampling loop
16:36 - U-Net
18:31 - Training code
19:28 - Sampling code
20:34 - Full code