Research project on hidden markov models for diffusion on images
This repository contained an uncleaned version of the code for "Diffusion priors in variational autoencoders" by Antoine Wehenkel and Gilles Louppe. It also contains some exploration on how diffusion models can be used to perform semantic representation learning by using a DDPM to model the latent variables of a Hidden Markov Chain in which the observation space is noised in a way that relates to the semantic. If you make any use of this code please cite us with @article{wehenkel2021diffusion, title={Diffusion priors in variational autoencoders}, author={Wehenkel, Antoine and Louppe, Gilles}, journal={arXiv preprint arXiv:2106.15671}, year={2021} }