Are Pretrained Image Matchers Good Enough for SAR-Optical Satellite Registration?

Isaac Corley; Alex Stoken; Gabriele Berton

cs.CVApr 11, 2026

Are Pretrained Image Matchers Good Enough for SAR-Optical Satellite Registration?

Authors: Isaac Corley, Alex Stoken, Gabriele Berton

Organizations: Taylor Geospatial · Independent Researcher

Abstract

Cross-modal optical--SAR (Synthetic Aperture Radar) registration is a bottleneck in remote-sensing disaster response. Modern image matchers are developed and benchmarked almost exclusively on natural-image domains. We evaluate twenty-four pretrained matcher configurations in a zero-shot setting, with no fine-tuning or domain adaptation on satellite or SAR data. The evaluation spans SpaceNet9 and two additional cross-modal benchmarks under a deterministic protocol that uses tiled large-image inference, robust geometric filtering, and tie-point-grounded metrics. Our results show uneven transfer: matchers with explicit cross-modal training do not uniformly outperform those without it. XoFTR (trained for visible--thermal matching) and RoMa achieve the lowest reported mean tie-point error at $3.0$ px on the labeled SpaceNet9 training scenes. RoMa achieves this result without any cross-modal training. MatchAnything-ELoFTR ( $3.4$ px), trained on synthetic cross-modal pairs, is close behind. XoFTR also runs roughly an order of magnitude faster than the RoMa family ( $0.4$ s vs. $5.0$ s per pair). RoMa's performance is consistent with the hypothesis that frozen DINOv2 features confer robustness to large appearance shifts. Deployment protocol choices (geometry model, tile size, inlier gating) change mean error by up to $33\times$ for a single matcher. This shift can exceed the effect of swapping matchers within our evaluated protocol ablations. Affine geometry alone reduces mean error from $12.3$ to $9.7$ px.

Figures & tables

Dataset	Modalities	Supervision	Resolution	Coverage
SpaceNet9 [ 10 ]	Optical–SAR	Tie points	${\sim}$ 0.5 m	3 scenes
SRIF [ 17 ]	Opt–SAR/Opt–IR	Affine GT ( $2{\times}3$ )	Varies	600 pairs
SARptical [ 12 ]	SAR–Optical patches	Binary match/no-match	${\sim}$ 1 m	40 queries

Table 1: Dataset summary. Geometry-labeled sets (SpaceNet9, SRIF) support transform-grounded evaluation; the pair-only set (SARptical) supports matchability and retrieval metrics.

Dense warp (DINOv2/3 backbone)
Matcher	Norm	MeanErr $\downarrow$	S@5 $\uparrow$	S@10 $\uparrow$	Fail $\downarrow$
RoMa+Tiny-RoMa [ 7 ]	Percentile	3.6	66.6	83.8	0.00
RoMaV2 [ 8 ]	Z-Score	3.6	69.2	90.2	0.00
RoMa+LoFTR [ 7 , 35 ]	Percentile	3.3	75.9	90.1	0.00
RoMa [ 7 ]	Z-Score	3.0	78.9	94.2	0.00
Cross-modal / multimodal-pretrained

Table 2: SpaceNet9: Top-15 matchers (best norm per matcher). All 24 matchers evaluated under a single protocol: tiled inference (512 px tiles, 256 px overlap), affine geometry, 1024 px max side, RANSAC threshold 3.0 px, min 4 inliers. Each row shows the best normalization for that matcher. MeanErr in pixels; S@

k

in %; Fail in [0,1].

Matcher	Norm	MeanErr (px) $\downarrow$	S@5 $\uparrow$	S@10 $\uparrow$	Fail $\downarrow$
SuperPoint-LightGlue [ 18 ]	CLAHE	67.0	31.3	39.0	0.64
RoMa [ 7 ]	CLAHE	64.9	36.0	40.4	0.00
RoMa+LoFTR [ 7 ]	CLAHE	64.6	36.1	40.6	0.00
RoMa+Tiny-RoMa [ 7 ]	CLAHE	63.9	35.3	40.0	0.00
LoFTR [ 35 ]	CLAHE	63.3	29.7	39.4	0.65
LoFTR [ 35 ]	Z-Score	63.1	30.6	41.7	0.64

Table 3: SRIF: Normalization ablation (600 pairs). Top matchers from SpaceNet9 evaluated on SRIF across all three normalizations, showing sensitivity to preprocessing choice.

Matcher	Norm	AUROC $\uparrow$	AUPRC $\uparrow$	R@1 $\uparrow$	R@5 $\uparrow$
Tiny-RoMa [ 7 ]	CLAHE	0.52	0.10	7.5	35.0
SuperPoint-LightGlue [ 18 ]	Z-Score	0.53	0.09	67.5	100.0
DISK-LightGlue [ 18 ]	Identity	0.53	0.09	70.0	97.5
LoFTR [ 35 ]	Identity	0.54	0.10	10.0	50.0
HardNet-LightGlue [ 20 ]	Identity	0.54	0.11	25.0	50.0
DUSt3R [ 41 ]	CLAHE	0.55	0.10	92.5	100.0

Table 4: SARptical: Top-10 configurations by AUROC. Each configuration uses 40 SAR queries over 525 candidate pairs; AUROC/AUPRC measure pair-level ranking quality over the full candidate pool, while Recall@

K

measures shortlist hit rate; these metrics can diverge.

Figure 2: Protocol sensitivity per matcher (SpaceNet9). Y-axis: mean error (px,

\downarrow

better). Each data point is one result from the threshold-robustness, keypoint-budget, or inlier-gating ablations, aggregated across seven matchers (MA-ELoFTR, MINIMA-RoMa, RoMa, XoFTR, LoFTR, MINIMA-XoFTR, XFeat; 133 total runs). Intra -matcher variance often matches or exceeds inter -matcher differences. Protocol choices can matter as much as matcher selection in the evaluated ablations.

Panel of Figure 3 — Figure 3: Core protocol effects on SpaceNet9. Geometry choice and tiling interact strongly with matcher performance.

Figure 4: RANSAC threshold robustness. S@10 (

\uparrow

better) as a function of reprojection threshold (px) for eight matchers under fixed tiled affine settings. MA-ELoFTR and MINIMA-RoMa maintain

>

90% success at permissive thresholds; XoFTR and RoMa follow closely. Sparser or less robust matchers (LoFTR, XFeat) are more threshold-sensitive. Their deployment accuracy depends strongly on threshold selection.

Figure 5: Normalization $\times$ matcher heatmap (SRIF). Cell values are mean corner reprojection error (px); greater column-wise spread indicates greater sensitivity to the normalization scheme. MINIMA-RoMa with Z-Score achieves the lowest error.

Panel of Figure 6 — Figure 6: SpaceNet9 qualitative correspondence gallery. Panels show top matchers under their best zero-shot configurations; green lines denote affine-RANSAC inliers and orange lines matched outliers.

Matcher	T (s)	Matcher	T (s)
XFeat-Star	0.3	DeDoDe-LG	3.8
XFeat	0.3	RoMaV2	4.1
Tiny-RoMa	0.3	MA-ELoFTR	4.7
MINIMA-RoMa-Tiny	0.4	DUSt3R	4.9
XoFTR	0.4	RoMa	5.0
SIFT-NN	0.4	MINIMA-XoFTR	5.3

Table 5: Per-pair runtime on SpaceNet9 (RTX 3090). End-to-end wall-clock per pair (tile extraction, matcher forward pass, and RANSAC fit) under the Table 2 protocol (1024 px max side,

512{\times}512

tiles with 256 px overlap,

\sim

4 tiles per pair), averaged over all labeled SpaceNet9 pairs. Sorted by time within each column; matchers span more than an order of magnitude. Preprocessing and I/O are excluded as they are essentially constant across matchers.

Cross-Modal Image Registration

Cross-modal image registration aligns images from different sensors or modalities into a shared spatial frame.

See all 67 papers

TAR: Text Semantic Assisted Cross-modal Image Registration Framework for Optical and SAR Images
May 12, 2026Zhuoyu Cai, Dou Quan, Ning Huyan +3Remote Sensing Synthetic Aperture Radar
Key Laboratory of Intelligent Perception and Image Understanding of Ministry of Education of China, School of Artificial Intelligence, Xidian University, Xi’an 710071, China
Preserving the Manifold: Frozen Expert Alignment for Label-Efficient SAR Scene Understanding
Oct 18, 2026Eli A. Underwood, I. Taylor McKechnie, Leslie M. CollinsRemote Sensing VLMs Cross-Modal Alignment
Duke University
Cross-modal learning for SAR target recognition using optical vision foundation models
Sep 7, 2026Lucas Hirsch, James R. Hopgood, Javid Khan +2Cross-Modal Alignment Cross-Modal Representation Transfer
School of Engineering, University of Edinburgh, United Kingdom · Leonardo UK, United Kingdom · School of Engineering and Physical Sciences, Heriot-Watt University, United Kingdom

May 12, 2026J/K move · Enter open · S save

TAR: Text Semantic Assisted Cross-modal Image Registration Framework for Optical and SAR Images

Zhuoyu Cai, Dou Quan, Ning Huyan, Pei He+2

Key Laboratory of Intelligent Perception and Image Understanding of Ministry of Education of China, School of Artificial Intelligence, Xidian University, Xi’an 710071, China

Remote Sensing Synthetic Aperture Radar Cross-Modal Feature Matching Cross-Modal Image Registration Deformable Image Registration

Open paper AskPDF arXiv

Existing deep learning-based methods can capture shared features from optical and synthetic aperture radar (SAR) images for spatial alignment. However, optical-SAR registration remains challenging under large geometric deformations, because the model needs to simultaneously handle cross-modal appearance discrepancies and complex spatial transformations. To address this issue, this paper proposes a text semantic-assisted cross-modal image registration framework, named TAR, for optical and SAR images. TAR exploits text semantic priors from remote sensing scenes and land-cover categories to alleviate the modality gap and enhance cross-modal feature learning. TAR consists of three components: a multi-scale visual feature learning (MSFL) module, a text-assisted feature enhancement (TAFE) module, and a coarse-to-fine dense matching (CFDM) module. MSFL extracts multi-scale visual features from optical and SAR images. TAFE constructs text descriptors related to remote sensing scenes and land-cover objects, and uses a frozen RemoteCLIP text encoder to extract text features. These text features are introduced through visual-text interaction to enhance high-level visual features for more reliable coarse matching. CFDM then establishes coarse correspondences based on the enhanced high-level features and refines the matched locations using low-level features. Experimental results on cross-modal remote sensing images demonstrate the effectiveness of TAR, which achieves stronger matching performance than several state-of-the-art methods and yields significant gains under large geometric deformations.

Abstract

Cross-Modal Image Registration

Explore similar work

TAR: Text Semantic Assisted Cross-modal Image Registration Framework for Optical and SAR Images

Preserving the Manifold: Frozen Expert Alignment for Label-Efficient SAR Scene Understanding

Cross-modal learning for SAR target recognition using optical vision foundation models