Saving Foundation Flow-Matching Priors for Inverse Problems

2025-11-20

AI Summary

To address the limited generalization and task adaptability of foundational flow matching (FM) models in inverse problems (IPs), this paper introduces FMPlug—a plug-and-play framework for diverse image restoration and scientific IPs. Methodologically, FMPlug features: (1) an instance-guided time-varying warm-start strategy for task-specific dynamic initialization; (2) sharp Gaussian regularization to preserve statistically well-structured solution manifolds without compromising model generality; and (3) a modular, plugin-style architecture enabling zero-shot transfer across heterogeneous IP tasks. Extensive experiments demonstrate that FMPlug consistently outperforms both existing FM-based baselines and task-specific models across multiple IPs—including deblurring, super-resolution, and tomographic reconstruction. To our knowledge, this is the first work to empirically validate foundational FM models as reusable, plug-and-play universal priors—establishing their feasibility and superiority over specialized alternatives.

Technology Category

• Computer Vision: Other Foundations of Computer Vision
• Machine Learning: Deep Neural Architectures and Foundation Models
• Planning, Routing, and Scheduling: Replanning and Plan Repair

Application Category

• User Modeling, Personalization and Recommendation: Federated recommendation systems and personalization
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Foundation flow-matching (FM) models promise a universal prior for solving inverse problems (IPs), yet today they trail behind domain-specific or even untrained priors. How can we unlock their potential? We introduce FMPlug, a plug-in framework that redefines how foundation FMs are used in IPs. FMPlug combines an instance-guided, time-dependent warm-start strategy with a sharp Gaussianity regularization, adding problem-specific guidance while preserving the Gaussian structures. This leads to a significant performance boost across image restoration and scientific IPs. Our results point to a path for making foundation FM models practical, reusable priors for IP solving.

Problem

• Unlocking foundation flow-matching models' potential for inverse problems
• Combining instance-guided warm-start with Gaussianity regularization strategy
• Improving performance across image restoration and scientific applications

Innovation

• Plug-in framework redefines foundation flow-matching usage
• Combines time-dependent warm-start with Gaussianity regularization
• Preserves Gaussian structures while adding problem-specific guidance