One-Step Face Restoration via Shortcut-Enhanced Coupling Flow

2026-03-03

AI Summary

This work addresses the limitations of existing flow-matching-based face restoration methods, which initialize from Gaussian noise and neglect the intrinsic dependency between low-quality (LQ) and high-quality (HQ) images, leading to curved trajectories, path crossings, and the need for multi-step sampling. To overcome these issues, the authors propose a data-dependent coupling mechanism that explicitly models the LQ–HQ relationship through conditional mean estimation and introduces a shortcut path constraint to supervise the average velocity field at any time step. This approach achieves high-fidelity single-step face restoration within the flow-matching framework for the first time, significantly improving restoration quality while maintaining inference speed comparable to conventional non-diffusion methods, thereby establishing state-of-the-art performance among single-step techniques.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Learning with Manifolds
• Planning, Routing, and Scheduling: Replanning and Plan Repair

Application Category

• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Face restoration has advanced significantly with generative models like diffusion models and flow matching (FM), which learn continuous-time mappings between distributions. However, existing FM-based approaches often start from Gaussian noise, ignoring the inherent dependency between low-quality (LQ) and high-quality (HQ) data, resulting in path crossovers, curved trajectories, and multi-step sampling requirements. To address these issues, we propose Shortcut-enhanced Coupling flow for Face Restoration (SCFlowFR). First, it establishes a \textit{data-dependent coupling} that explicitly models the LQ--HQ dependency, minimizing path crossovers and promoting near-linear transport. Second, we employ conditional mean estimation to obtain a coarse prediction that refines the source anchor to tighten coupling and conditions the velocity field to stabilize large-step updates. Third, a shortcut constraint supervises average velocities over arbitrary time intervals, enabling accurate one-step inference. Experiments demonstrate that SCFlowFR achieves state-of-the-art one-step face restoration quality with inference speed comparable to traditional non-diffusion methods.

Problem

• face restoration
• flow matching
• low-quality to high-quality dependency
• one-step inference
• path crossovers

Innovation

• flow matching
• face restoration
• one-step inference
• data-dependent coupling
• shortcut constraint