A Gradient Flow Approach to Solving Inverse Problems with Latent Diffusion Models
By: Tim Y. J. Wang, O. Deniz Akyildiz
Potential Business Impact:
Lets computers fix blurry pictures perfectly.
Solving ill-posed inverse problems requires powerful and flexible priors. We propose leveraging pretrained latent diffusion models for this task through a new training-free approach, termed Diffusion-regularized Wasserstein Gradient Flow (DWGF). Specifically, we formulate the posterior sampling problem as a regularized Wasserstein gradient flow of the Kullback-Leibler divergence in the latent space. We demonstrate the performance of our method on standard benchmarks using StableDiffusion (Rombach et al., 2022) as the prior.
Similar Papers
Diffusion prior as a direct regularization term for FWI
Geophysics
Makes earthquake maps clearer and faster.
Latent Refinement via Flow Matching for Training-free Linear Inverse Problem Solving
CV and Pattern Recognition
Makes blurry pictures clear using smart math.
Unifying and extending Diffusion Models through PDEs for solving Inverse Problems
Machine Learning (CS)
New math helps computers create realistic images.