Deep S2P: Integrating Learning Based Stereo Matching Into the Satellite Stereo Pipeline

Elías Masquil; Thibaud Ehret; Pablo Musé; Gabriele Facciolo

cs.CVMar 23, 2026

Deep S2P: Integrating Learning Based Stereo Matching Into the Satellite Stereo Pipeline

Authors: Elías Masquil, Thibaud Ehret, Pablo Musé, Gabriele Facciolo

Organizations: AMIAD, Pôle Recherche, France · Université Paris-Saclay, ENS Paris-Saclay, CNRS, Centre Borelli, 91190, Gif-sur-Yvette, France · Institut Universitaire de France

Abstract

Digital Surface Model generation from satellite imagery is a core task in Earth observation and is commonly addressed using classical stereoscopic matching algorithms in satellite pipelines as in the Satellite Stereo Pipeline (S2P). While recent learning-based stereo matchers achieve state-of-the-art performance on standard benchmarks, their integration into operational satellite pipelines remains challenging due to differences in viewing geometry and disparity assumptions. In this work, we integrate several modern learning-based stereo matchers, including StereoAnywhere, MonSter, Foundation Stereo, and a satellite fine-tuned variant of MonSter, into the Satellite Stereo Pipeline, adapting the rectification stage to enforce compatible disparity polarity and range. We release the corresponding code to enable reproducible use of these methods in large-scale Earth observation workflows. Experiments on satellite imagery show consistent improvements over classical cost-volume-based approaches in terms of Digital Surface Model accuracy, although commonly used metrics such as mean absolute error exhibit saturation effects. Qualitative results reveal substantially improved geometric detail and sharper structures, highlighting the need for evaluation strategies that better reflect perceptual and structural fidelity. At the same time, performance over challenging surface types such as vegetation remains limited across all evaluated models, indicating open challenges for learning-based stereo in natural environments.

Figures & tables

Method	P90	NMAD	RMSE	MAE	% Valid Points
S2P-HD (SGM)	$2.30\pm 1.86$	$1.03\pm 0.35$	$5.19\pm 4.16$	$2.25\pm 0.87$	$78.51\pm 19.57$
S2P-HD (MGM)	$2.82\pm 3.91$	$1.13\pm 0.74$	$6.15\pm 6.47$	$3.07\pm 2.86$	$81.15\pm 15.44$
Diachronic Stereo	$3.22\pm 2.60$	$1.21\pm 0.95$	$3.66\pm 2.11$	$2.28\pm 1.22$	$90.89\pm 19.29$
FoundationStereo	$1.91\pm 2.23$	$0.53\pm 0.26$	$3.32\pm 1.67$	$1.96\pm 0.92$	$88.51\pm 8.80$
MonSter	$2.06\pm 2.20$	$0.69\pm 0.33$	$3.37\pm 1.74$	$2.05\pm 0.93$	$87.58\pm 18.44$
StereoAnywhere	$2.70\pm 2.70$	$0.99\pm 0.63$	$3.59\pm 1.79$	$2.20\pm 1.02$	$90.45\pm 12.73$

TABLE I: Performance comparison on stereo pairs from the GRSS dataset. Best results are shown in bold , second best are underlined .

Method	Class	P90	NMAD	RMSE	MAE
S2P-HD (SGM)	Ground	$1.35\pm 1.30$	$0.83\pm 0.35$	$4.10\pm 4.69$	$1.85\pm 0.90$
S2P-HD (MGM)	Ground	$1.44\pm 1.32$	$0.81\pm 0.38$	$4.01\pm 3.43$	$2.04\pm 1.08$
Diachronic Stereo	Ground	$1.27\pm 2.00$	$0.85\pm 0.62$	$2.44\pm 1.49$	$1.67\pm 0.97$
FoundationStereo	Ground	$0.78\pm 1.33$	$0.40\pm 0.15$	$2.37\pm 1.36$	$1.52\pm 0.79$
MonSter	Ground	$0.79\pm 1.37$	$0.48\pm 0.23$	$2.40\pm 1.47$	$1.61\pm 0.94$
StereoAnywhere	Ground	$1.07\pm 1.41$	$0.65\pm 0.23$	$2.52\pm 1.40$	$1.65\pm 0.75$

TABLE II: Semantic-wise DSM error analysis on the GRSS dataset (best pairs). Best results are shown in bold , second best are underlined .

Method	P90	NMAD	RMSE	MAE	% Valid Points
Diachronic Stereo	$2.10\pm 2.42$	$0.44\pm 0.59$	$2.63\pm 1.66$	$1.24\pm 1.20$	$65.72\pm 21.77$
FoundationStereo	$2.03\pm 2.43$	$0.39\pm 0.60$	$2.85\pm 1.91$	$1.44\pm 1.33$	$64.85\pm 17.87$
MonSter	$4.19\pm 4.25$	$0.53\pm 0.78$	$3.75\pm 2.10$	$1.89\pm 1.51$	$61.41\pm 26.87$
StereoAnywhere	$3.43\pm 4.49$	$0.63\pm 1.10$	$3.60\pm 2.96$	$1.81\pm 2.23$	$61.13\pm 15.48$

TABLE III: Performance on challenging stereo pairs from the GRSS dataset. Best results are shown in bold , second best are underlined .

Panel of Figure 4 — Fig. 1: DSM reconstructions for two AOIs. Top row: JAX_068 (urban scene). Bottom row: JAX_559 (suburban scene with dense vegetation).

Stereo Matching

Stereo matching estimates depth by finding corresponding points in paired images and measuring their disparity.

See all 57 papers

Jul 29, 2026

SeasonStereo: Robust Dense Stereo Matching for Multi-Date Satellite Imagery via Generative AI

Accurate 3D reconstruction from satellite imagery typically relies on near-simultaneous stereo pairs, limiting its applicability to diachronic settings where multi-date images exhibit varying seasonal and illumination conditions. Training dense stereo matching models robust to appearance changes is a long-standing challenge, as aligned multi-date imagery and ground-truth geometry are costly to obtain at scale. We propose SeasonStereo, a scalable framework that addresses disparity estimation from diachronic satellite images by training on synthetic image pairs with controlled seasonal appearance variation, while leveraging zero-shot geometric priors from foundation models. SeasonStereo matches the accuracy of state-of-the-art LiDAR-supervised models, while producing sharper geometric details without requiring aligned real multi-date training products or LiDAR-derived labels. As a result, SeasonStereo offers a practical path toward large-scale 3D reconstruction from heterogeneous satellite images with reduced supervision cost.

Álvaro Díaz-Laureano, Roger Marí, Elías Masquil +2

Eurecat, Multimedia Technologies, Barcelona, Spain · IIE, Facultad de Ingeniería, Universidad de la República, Uruguay · Dept. of Engineering, Universitat Pompeu Fabra, Barcelona, Spain +2

Remote Sensing Image Understanding 3D Reconstruction Stereo Matching

Ask PDF

May 8, 2026

Sat3R: Satellite DSM Reconstruction via RPC-Aware Depth Fine-tuning

Accurate Digital Surface Model (DSM) reconstruction from satellite imagery is critical for applications such as disaster response, urban planning, and large-scale geographic mapping. Existing approaches face a fundamental trade-off: optimization-based methods achieve strong accuracy but require hours of per-scene computation, while generalizable geometry foundation models offer near-instant inference but fail to generalize to satellite imagery due to the domain gap introduced by the Rational Polynomial Camera (RPC) model and mismatched depth scale distributions. We present Sat3R, a feed-forward framework that bridges this gap via RPC-aware metric depth fine-tuning of Depth Anything V2 using the Scale-Invariant Logarithmic (SiLog) loss. By constructing physically consistent pseudo depth supervision from RPC geometry, Sat3R adapts a monocular depth foundation model to the satellite domain without per-scene optimization. Experiments on the DFC2019 benchmark demonstrate that Sat3R reduces MAE by 38% over zero-shot feed-forward baselines and achieves competitive accuracy against optimization-based methods, while delivering over 300x speedup. Sat3R demonstrates that feed-forward models, when properly adapted to the satellite domain, can match optimization-based accuracy at a fraction of the computational cost, paving the way for practical large-scale satellite DSM reconstruction.

Qiaoyi Yang, Chaoyi Zhou, Xi Liu +9

1Clemson University · University of Washington · University of Tennessee

Metric Depth Estimation 3D Reconstruction Monocular Depth Estimation Remote Sensing Feed-Forward 3D Reconstruction

Ask PDF

Nov 19, 2024

Motif Channel Opened in a White-Box: Stereo Matching via Motif Correlation Graph

Real-world applications of stereo matching, such as autonomous driving, place stringent demands on both safety and accuracy. However, learning-based stereo matching methods inherently suffer from the loss of geometric structures in certain feature channels, creating a bottleneck in achieving precise detail matching. Additionally, these methods lack interpretability due to the black-box nature of deep learning. In this paper, we propose MoCha-V2, a novel learning-based paradigm for stereo matching. MoCha-V2 introduces the Motif Correlation Graph (MCG) to capture recurring textures, which are referred to as ``motifs" within feature channels. These motifs reconstruct geometric structures and are learned in a more interpretable way. Subsequently, we integrate features from multiple frequency domains through wavelet inverse transformation. The resulting motif features are utilized to restore geometric structures in the stereo matching process. Experimental results demonstrate the effectiveness of MoCha-V2. MoCha-V2 achieved 1st place on the Middlebury benchmark at the time of its release. Code is available at https://github.com/ZYangChen/MoCha-Stereo.

Ziyang Chen, Yongjun Zhang, Wenting Li +3

College of Computer Science, the State Key Laboratory of Public Big Data, Guizhou University, Guiyang 550025, China · School of Information Engineering, Guizhou University of Commerce, Guiyang 550021, China · School of Software, Northwestern Polytechnical University, Xi’an, 710129, China +2

Representation Learning Stereo Matching

Ask PDF

Abstract

Stereo Matching

Explore similar work

SeasonStereo: Robust Dense Stereo Matching for Multi-Date Satellite Imagery via Generative AI

Sat3R: Satellite DSM Reconstruction via RPC-Aware Depth Fine-tuning

Motif Channel Opened in a White-Box: Stereo Matching via Motif Correlation Graph