MatRes: Zero-Shot Test-Time Model Adaptation for Simultaneous Matching and Restoration
This work addresses the mutual interference between image restoration and geometric matching in real-world image pairs with severe degradations and viewpoint changes, offering a practical solution for users capturing multiple images of a scene.
MatRes is a zero-shot test-time adaptation framework that jointly improves image restoration and geometric matching using a single low-quality and high-quality image pair, yielding significant gains in both tasks compared to using either restoration or matching models alone.
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.