Diffusion models for generating images and Pyro

Dear Pyro Forum,

I was wondering what the state of diffusion models and Pyro are. Is this an area that is actively being worked on at the moment?

Currently I am working on generating images using a deep Bayesian net where the images are generating using a hierarchical VAE, but the images are simply not sharp enough. This is where diffusion models come in, but I am a bit worried with respect to Rao–Blackwellization and other tricks used in training this class of models. I was just wondering if anyone with more experience than me thinks that Pyro is the appropriate tool for building very deep diffusion models for generating images?

you could write a diffusion model in pyro but i’m not sure you’d get much benefit from doing so as compared to writing such a model in pure pytorch