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

Kanggeon Lee; Soochahn Lee; Kyoung Mu Lee

cs.CVApr 11, 2026

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

Authors: Kanggeon Lee, Soochahn Lee, Kyoung Mu Lee

Organizations: ASRI, Dept. of ECE, Seoul National University, Korea · Dept. of Electronics Engineering, Kookmin University, Korea

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.

Figures & tables

Figure 2: Method Overview. Given an input pair $(\mathcal{I}_{\texttt{LQ}},\mathcal{I}_{\texttt{HQ}})$ , the pretrained generative prior model $\mathcal{M}_{\theta}$ and the pretrained restoration network $\mathcal{R}_{\phi}$ remain frozen during adaptation (gradient flows through both models but their parameters are not updated). The zero-initialized adapter $\mathcal{H}_{\psi}$ is the only trainable module, refined iteratively using losses computed on the $\mathcal{I}_{\texttt{EQ}}$ and $\mathcal{I}_{\texttt{HQ}}$ image pairs. At each adaptation step, $\mathcal{M}_{\theta}$ extracts generative priors and estimates the transformation $\mathcal{T}_{\texttt{LQ}\rightarrow\texttt{HQ}}$ , the $\mathcal{I}_{\texttt{LQ}}$ image are warped and injected into $\mathcal{R}_{\phi}$ , and the adapter $\mathcal{H}_{\psi}$ learns to align and modulate features for improved restoration.

Figure 3: Qualitative Restoration Results. Restoration performance is compared with and without the matching network $\mathcal{M}_{\theta}$ across three tasks. For super-resolution (SR), we evaluate EDSR [ 33 ] , ESRGAN [ 45 ] , SwinIR [ 32 ] , and HAT [ 46 ] ; for denoising, we evaluate DnCNN [ 47 ] , N2V [ 35 ] , N2N [ 34 ] , and Restormer [ 31 ] ; and for deblurring, we evaluate DeblurGAN [ 36 ] , SRN [ 37 ] , Restormer [ 31 ] , and NAFNet [ 48 ] . The green box indicates the zoomed-in region.

Figure 4: Qualitative Matching Results. Geometric transform estimation results for five matching networks (LoftUp [ 26 ] , ART [ 27 ] , DIFT [ 28 ] , RoMa [ 29 ] , Mast3R [ 30 ] ) across SR, denoising, and deblurring tasks, is evaluated with and without the restoration network $\mathcal{R}_{\phi}$ . The green box denotes the ground-truth transform, while the red box represents the estimated transform.

Task	Matcher ( $\mathcal{M}$ )	Restorator ( $\mathcal{R}$ )	Matching score		Restoration score
Task	Matcher ( $\mathcal{M}$ )	Restorator ( $\mathcal{R}$ )	mAUC (w/o $\mathcal{R}$ )	mAUC (w/ $\mathcal{R}$ )	PSNR (w/o $\mathcal{M}$ )	PSNR (w/ $\mathcal{M}$ )	SSIM (w/o $\mathcal{M}$ )	SSIM (w/ $\mathcal{M}$ )
SR	RoMa [ 29 ]	EDSR [ 33 ]	$27.00$	$36.60$ ( $+9.61$ )	$26.93$	$31.72$ ( $+4.80$ )	$0.531$	$0.747$ ( $+0.216$ )
		ESRGAN [ 45 ]	$24.46$	$33.39$ ( $+8.93$ )	$26.40$	$31.53$ ( $+5.12$ )	$0.505$	$0.823$ ( $+0.317$ )
		SwinIR [ 32 ]	$30.66$	$34.36$ ( $+3.70$ )	$24.73$	$28.28$ ( $+3.55$ )	$0.497$	$0.688$ ( $+0.191$ )
		HAT [ 46 ]	$27.46$	$31.79$ ( $+4.33$ )	$26.45$	$29.87$ ( $+3.42$ )	$0.534$	$0.716$ ( $+0.183$ )
	ART [ 27 ]	EDSR [ 33 ]	$27.65$	$35.93$ ( $+8.28$ )	$24.80$	$29.34$ ( $+4.54$ )	$0.527$	$0.670$ ( $+0.143$ )

Table 1: Experimental Results for Super-Resolution, Denoising, and Deblurring Tasks

Task	Dataset	Restoration Score (PSNR / SSIM)
Methods:		MatRes (ours)	AutoLUT [ 59 ]	CATANet [ 60 ]	IM-LUT [ 61 ]
SR	RealSR [ 49 ]	$\mathbf{34.87}$ / $\mathbf{0.902}$	$33.59$ / $0.864$	$32.70$ / $0.839$	$30.66$ / $0.777$
	DIV2K [ 50 ]	$32.78$ / $0.758$	$31.86$ / $0.733$	$\mathbf{34.72}$ / $\mathbf{0.801}$	$32.04$ / $0.700$
	Set5 [ 51 ]	$33.31$ / $0.742$	$\mathbf{33.98}$ / $\mathbf{0.779}$	$31.51$ / $0.700$	$30.39$ / $0.700$
	Set14 [ 52 ]	$35.01$ / $0.941$	$34.02$ / $0.917$	$33.54$ / $0.899$	$\mathbf{35.86}$ / $\mathbf{0.962}$
	Urban100 [ 53 ]	$\mathbf{35.82}$ / $\mathbf{0.776}$	$34.23$ / $0.758$	$31.69$ / $0.700$	$31.80$ / $0.732$

Table 2: Restoration score comparison across different tasks.

Task	Dataset	Matching Score (mAUC)
Task	Dataset	MatRes (ours)	MINIMA [ 68 ]	DenseAffine [ 69 ]	MatchAnything [ 70 ]
SR	RealSR [ 49 ]	$\mathbf{34.85}$	$24.99$	$26.06$	$29.60$
	DIV2K [ 50 ]	$\mathbf{32.07}$	$21.41$	$26.38$	$26.62$
	Set5 [ 51 ]	$\mathbf{31.38}$	$23.18$	$23.92$	$25.05$
	Set14 [ 52 ]	$\mathbf{34.31}$	$27.98$	$27.38$	$26.66$
	Urban100 [ 53 ]	$\mathbf{35.88}$	$27.77$	$27.14$	$31.51$

Table 3: Matching score comparison across different tasks.

Image Matching

Image matching finds corresponding points, regions, or objects across images despite changes in viewpoint, scale, lighting, or sensor.

See all 87 papers

Generative Manifold Distillation: Aligning Restoration Trajectories with Natural Image Prior
Dec 11, 2025Yuyang Hu, Mojtaba Sahraee-Ardakan, Arpit Bansal +4Source-Free Domain Adaptation Image Restoration
Google · Washington University in St. Louis
Self-supervised Dynamic Heterogeneous Degradation Modeling for Unified Zero-Shot Image Restoration
May 23, 2026XiaoWan Hu, Jing Yang, HeNan Liu +2Diffusion Models for Image Restoration Diffusion Model Sampling
Beihang University · Tsinghua University
Beyond Spatio-Temporal Priors: A Generalizable Approach for Dense Correspondence Matching
Oct 8, 2026Luping Liu, Bingyi Kang, Yifan Wang +1Image Matching
The University of Hong Kong · ByteDance Seed · Zhejiang University

Dec 11, 2025J/K move · Enter open · S save

Generative Manifold Distillation: Aligning Restoration Trajectories with Natural Image Prior

Yuyang Hu, Mojtaba Sahraee-Ardakan, Arpit Bansal, Kangfu Mei+3

Google · Washington University in St. Louis

Source-Free Domain Adaptation Image Restoration Knowledge Distillation

Open paper AskPDF arXiv

Pre-trained image restoration models often fail on out-of-distribution (OOD) real-world degradations. Adapting to these domains is challenging as real-world data lacks paired ground truth, and unsupervised methods often require unstable architectural changes. We propose Generative Manifold Distillation (GMD), which reframes domain adaptation as geometric manifold alignment. GMD operates in a strictly unpaired setting, requiring only low-quality (LQ) target observations. By leveraging the flow-matching dynamics of a frozen text-to-image foundation model, GMD projects off-manifold restorations onto the natural image manifold to generate high-quality pseudo-targets. To ensure stability, a quality-gated manifold filter rejects off-manifold samples, while source-anchored trajectory regularization prevents error accumulation. Ultimately, GMD distills a powerful generative prior into an efficient restoration network. Experiments demonstrate that GMD seamlessly adapts to new distributions using only LQ inputs, drastically improving perceptual quality with zero architectural modifications or added inference latency.

Abstract

Image Matching

Explore similar work

Generative Manifold Distillation: Aligning Restoration Trajectories with Natural Image Prior

Self-supervised Dynamic Heterogeneous Degradation Modeling for Unified Zero-Shot Image Restoration

Beyond Spatio-Temporal Priors: A Generalizable Approach for Dense Correspondence Matching