SP$^3$: Spherical Priors for Plug-and-Play Restoration

2026-06-15

AI Summary

This work addresses the computational inefficiency and limited real-time applicability of conventional Plug-and-Play (PnP) image restoration methods, which rely on iterative denoisers. The authors propose SP³, a novel algorithm that introduces a spherical encoder into the PnP framework for the first time. SP³ replaces the proximal prior step with manifold projection onto the structured latent space of the spherical encoder and incorporates half-quadratic splitting to enable closed-form data-consistency updates. Notably, it achieves stable convergence without requiring gradient computation during inference. Being training-free and gradient-independent, SP³ offers an out-of-the-box solution that attains perceptual quality comparable to state-of-the-art zero-shot diffusion and flow-based models across diverse image restoration tasks, while accelerating inference by 3× to 630×.

Abstract

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 $3$-$630\times$ faster.

Problem

• image restoration
• Plug-and-Play
• generative priors
• maximum a posteriori
• efficient inference

Innovation

• Spherical Encoder
• Plug-and-Play
• Anytime Restoration
• Generative Prior
• Half-Quadratic Splitting