Summary
In this chapter, we took a deep dive into the internal workings of the diffusion model initially brought out by Jonathan Ho et al. [4]. We learned about the foundational ideas of the diffusion model and learned about the forward diffusion process. We also walked through the reverse diffusion process for diffusion model training and sampling and explored how to enable a text-guided diffusion model.
Through this chapter, we aimed to explain the core idea of the diffusion model. If you want to implement a diffusion model by yourself, I would recommend reading through the original DDPM paper directly.
The DDPM diffusion model can generate realistic images, but one of its problems is its performance. Not only is training a model slow, but the image sampling is also slow. In the next chapter, we are going to discuss the Stable Diffusion model, which will boost the speed in a genius way.
