MatRes: Zero-Shot Test-Time Model Adaptation for Simultaneous Matching and Restoration

TL;DR

MatRes is a zero-shot, test-time adaptation framework that jointly enhances image restoration and geometric matching from a single low-quality and high-quality image pair without additional training.

Contribution

It introduces a novel method that updates only lightweight modules for simultaneous restoration and matching, avoiding offline training and extra supervision.

Findings

Significant improvements in restoration quality and geometric alignment.

Effective across diverse image degradation and viewpoint change scenarios.

No need for offline training or additional supervision.

Abstract

Real-world image pairs often exhibit both severe degradations and large viewpoint changes, making image restoration and geometric matching mutually interfering tasks when treated independently. In this work, we propose MatRes, a zero-shot test-time adaptation framework that jointly improves restoration quality and correspondence estimation using only a single low-quality and high-quality image pair. By enforcing conditional similarity at corresponding locations, MatRes updates only lightweight modules while keeping all pretrained components frozen, requiring no offline training or additional supervision. Extensive experiments across diverse combinations show that MatRes yields significant gains in both restoration and geometric alignment compared to using either restoration or matching models alone. MatRes offers a practical and widely applicable solution for real-world scenarios where users commonly capture multiple images of a scene with varying viewpoints and quality, effectively addressing the often-overlooked mutual interference between matching and restoration.