Search papers, labs, and topics across Lattice.
This paper introduces SP^3, a novel Plug-and-Play algorithm that enhances maximum a posteriori image restoration by integrating Spherical Encoders (SE) as generative priors. By leveraging the SE's structured latent space for robust projections onto the natural image manifold, SP^3 achieves stable convergence through an alternating projection and closed-form data-consistency step, eliminating the need for gradient computation during inference. Evaluations show that SP^3 not only matches the perceptual quality of leading zero-shot diffusion and flow methods but does so at speeds 3 to 630 times faster.
Achieving state-of-the-art image restoration quality at unprecedented speeds, SP^3 redefines the efficiency of generative priors in practical applications.
In this paper, we introduce SP^3, a novel Plug-and-Play algorithm that accelerates maximum a posteriori image restoration by replacing denoisers with Spherical Encoders (SE) as generative priors. SP^3 approximates the intractable proximal prior step by utilizing the SE tightly structured latent space as a robust projection onto the natural image manifold. Alternating this projection with a closed-form data-consistency step, via Half-Quadratic Splitting, achieves stable convergence without requiring gradient computation during inference. This unique formulation unlocks "anytime" restoration capabilities, producing sharp, plausible images from the first iteration. Evaluations across a variety of image restoration tasks demonstrate that SP^3 achieves perceptual quality comparable to state-of-the-art zero-shot diffusion and flow methods while being 3-630times faster.
