Regularization by denoising diffusion process meets deep relaxation in phase

Eunju Cha

Abstract: Highlights•Introduces a novel PhaseCut regularization technique based on convex relaxation of phase for Fourier phase retrieval problem.•Adopts a variational sampler using pre-trained diffusion models for Fourier phase retrieval problem as image priors.•Achieves the state-of-the-art performance with various datasets of 256 × 256 images.