Fast and Stable Diffusion Inverse Solver with History Gradient Update

arXiv:2307.12070v2 Announce Type: replace
Abstract: Diffusion models have recently been recognised as efficient inverse problem solvers due to their ability to produce high-quality reconstruction results without relying on pairwise data training. Existing diffusion-based solvers utilize Gradient Descent strategy to get a optimal sample solution. However, these solvers only calculate the current gradient and have not utilized any history information of sampling process, thus resulting in unstable optimization progresses and suboptimal solutions. To address this issue, we propose to utilize the …

abstract arxiv cs.cv data diffusion diffusion models gradient history however quality results sample solution solver stable diffusion strategy training type update