Image Reconstruction
Image reconstruction aims to recover high-quality images from incomplete or noisy measurements, a crucial task across diverse scientific fields. Current research heavily utilizes deep learning, employing architectures like UNets, Vision Transformers, and diffusion models, often incorporating physics-based constraints or learned regularizers to improve reconstruction fidelity and efficiency. These advancements are significantly impacting various applications, including medical imaging (e.g., MRI, CT, PET, photoacoustic tomography), materials science, and astronomical imaging, by enabling faster scans, higher resolution, and improved diagnostic accuracy. The field is also actively exploring self-supervised learning and parameter-efficient fine-tuning to address data scarcity and computational limitations.
Papers
DEFT: Efficient Fine-Tuning of Diffusion Models by Learning the Generalised $h$-transform
Alexander Denker, Francisco Vargas, Shreyas Padhy, Kieran Didi, Simon Mathis, Vincent Dutordoir, Riccardo Barbano, Emile Mathieu, Urszula Julia Komorowska, Pietro Lio
Enhancing Dynamic CT Image Reconstruction with Neural Fields and Optical Flow
Pablo Arratia, Matthias Ehrhardt, Lisa Kreusser