Restoration and Diffusion-Bridge Priors
Turning pre-trained restoration networks into generative priors
Pioneered a new class of plug-and-play priors by generalizing Regularization-by-Denoising (RED) to leverage any pre-trained image restoration network. Developed a suite of algorithms (DRP, ShaRP, ADOBI) for image recovery using deterministic priors, stochastic priors, and diffusion bridge models. ICLR 2024, ICML 2025, IEEE TCI 2026.