Update: Blog posts: DDPMs from scratch
Diffusion models are a class of likelihood-based generative models that recently have been used to produce very high quality images compared to other existing generative models like GANs. For example, take a look at the latest research Imagen or GLIDEwhere the authors used diffusion models to generate very high quality images.
Although you can find a lot of material online regarding other generative models like GANs to learn from, the list of resources for learning about diffusion models is still sparse. On top of it, the mathematics behind the diffusion models is a bit harder to understand. To address this, we are creating this repo to give you enough material to make you understand the working of diffusion models and the maths involved.
We try to keep everything organized in notebooks which you can run on Colab. We are also organizing the content in a series of short blog-posts but that would take some time. Also, some of the notebooks presented here are marked as optional. These notebooks covers the theoretical parts that you should be aware of before reading about diffusion models.
| Chapter No | Topic |
Colab | GitHub |
|---|---|---|---|
| 1. | Random Variables (Optional) |  |
 |
| 2. | Gaussian Distribution and DDPMs | |
|
| 3. | A deep dive into DDPMs | |
|