Mode-Seeking for Inverse Problems with Diffusion Models
By: Sai Bharath Chandra Gutha, Ricardo Vinuesa, Hossein Azizpour
Potential Business Impact:
Fixes blurry pictures faster than before.
A pre-trained unconditional diffusion model, combined with posterior sampling or maximum a posteriori (MAP) estimation techniques, can solve arbitrary inverse problems without task-specific training or fine-tuning. However, existing posterior sampling and MAP estimation methods often rely on modeling approximations and can be computationally demanding. In this work, we propose the variational mode-seeking loss (VML), which, when minimized during each reverse diffusion step, guides the generated sample towards the MAP estimate. VML arises from a novel perspective of minimizing the Kullback-Leibler (KL) divergence between the diffusion posterior $p(\mathbf{x}_0|\mathbf{x}_t)$ and the measurement posterior $p(\mathbf{x}_0|\mathbf{y})$, where $\mathbf{y}$ denotes the measurement. Importantly, for linear inverse problems, VML can be analytically derived and need not be approximated. Based on further theoretical insights, we propose VML-MAP, an empirically effective algorithm for solving inverse problems, and validate its efficacy over existing methods in both performance and computational time, through extensive experiments on diverse image-restoration tasks across multiple datasets.
Similar Papers
Local MAP Sampling for Diffusion Models
Graphics
Improves image and science results by finding best answers.
Local MAP Sampling for Diffusion Models
Graphics
Makes blurry pictures sharp and clear again.
Diffusion models for inverse problems
Machine Learning (CS)
Makes blurry pictures clear using smart computer tricks.