The field of image reconstruction and denoising is witnessing significant advancements with the integration of deep learning techniques and traditional methods. Researchers are exploring the connection between Plug-and-Play methods and Denoising Diffusion Implicit Models, leading to improved reconstruction quality in single-pixel imaging tasks. The development of hybrid data-consistency modules and the analysis of various loss functions are also contributing to the advancement of image desmoking and reconstruction techniques. Furthermore, the investigation of the stability of recovery under noise in phase retrieval problems is providing new insights into the optimal recovery bounds. The introduction of diffusion models as powerful generative priors is also showing promising results in sparse-view CT reconstruction. Noteworthy papers include:

• Plug-and-play Diffusion Models for Image Compressive Sensing with Data Consistency Projection, which proposes a unified framework integrating learned priors with physical forward models.
• Near-Optimal Recovery Performance of PhaseLift for Phase Retrieval from Coded Diffraction Patterns, which provides a nearly optimal recovery bound for PhaseLift under adversarial noise.
• DICE: Diffusion Consensus Equilibrium for Sparse-view CT Reconstruction, which introduces a framework combining strong generative prior capabilities with measurement consistency.