Generative Image Prior

Generative image priors leverage deep generative models, such as diffusion models and score-based models, to incorporate prior knowledge about image statistics into various inverse problems. Current research focuses on improving the robustness and generalizability of these priors across different imaging modalities (MRI, EHT data) and data acquisition parameters, often employing techniques like phase augmentation and federated learning to address data limitations and privacy concerns. This approach offers significant advantages in image reconstruction tasks by enhancing image quality, reducing artifacts, and enabling uncertainty quantification, leading to improved accuracy and reliability in diverse applications ranging from medical imaging to astrophysics.