Principled Probabilistic Imaging using Diffusion Models as Plug-and-Play Priors

The black hole imaging problem is non-convex and highly ill-posed due to severe noise corruption and measurement sparsity.