Photogrammetric 3D reconstruction has long relied on traditional Structure-from-Motion (SfM) and Multi-View Stereo (MVS) methods, which provide high accuracy but face challenges in speed and scalability. Recently, learning-based MVS methods have emerged, aiming for faster and more efficient reconstruction. This work presents a comparative evaluation between a representative traditional MVS pipeline (COLMAP) and state-of-the-art learning-based approaches, including geometry-guided methods (MVSNet, PatchmatchNet, MVSAnywhere, MVSFormer++) and end-to-end frameworks (Stereo4D, FoundationStereo, DUSt3R, MASt3R, Fast3R, VGGT). Two experiments were conducted on different aerial scenarios. The first experiment used the MARS-LVIG dataset, where ground-truth 3D reconstruction was provided by LiDAR point clouds. The second experiment used a public scene from the Pix4D official website, with ground truth generated by Pix4Dmapper. We evaluated accuracy, coverage, and runtime across all methods. Experimental results show that although COLMAP can provide reliable and geometrically consistent reconstruction results, it requires more computation time. In cases where traditional methods fail in image registration, learning-based approaches exhibit stronger feature-matching capability and greater robustness. Geometry-guided methods usually require careful dataset preparation and often depend on camera pose or depth priors generated by COLMAP. End-to-end methods such as DUSt3R and VGGT achieve competitive accuracy and reasonable coverage while offering substantially faster reconstruction. However, they exhibit relatively large residuals in 3D reconstruction, particularly in challenging scenarios.
Figures & tables
Figure 1: Multi-View Stereo Methods.
Method
Number of Views
Characteristic
Category
COLMAP
N
Classical SfM+MVS pipeline with explicit geometry
Traditional
PatchmatchNet
N
Learned PatchMatch for cost volume–based depth regression
Geomery-Guided
MVSFormer++
N
Transformer-based cost aggregation with explicit geometry
Geomery-Guided
MVSAnywhere
N
Cost Volume Patchifier with view-independent and scale-independent geometry
Geomery-Guided
FoundationStereo
2
Zero-shot generalization
End-to-end
Stereo4D
2
Monocular priors + 4D cost volume with temporal consistency
End-to-end
Table 1: Overview and categorization of representative MVS methods considered in this study
Table 2: Datasets used for our tests.
Figure 2: Failed results on Island. The notation “Method–N” (e.g., MASt3R-50, VGGT-50) indicates the reconstruction result obtained from different image samples (N). The red circle marks a local area with discontinuities and layering effects in the island reconstructed point cloud. The zoomed-in view above highlights these artifacts.
Figure 3: Failed results on Urban. The notation “Method–N” (e.g., MASt3R-50, VGGT-50) indicates the reconstruction result obtained from different image samples (N). The red circle marks a local area with discontinuities and layering effects in the urban reconstructed point cloud. The zoomed-in view above highlights these artifacts.
Methods
2
10
20
50
100
DUSt3R
3.15
35.91
120.16
MASt3R
8.60
39.49
68.63
1075.49
VGGT
1.52
6.10
16.59
28.36
60.79
Fast3R
0.41
0.82
1.70
5.57
16.47
MVSAnywhere
N/A
N/A
N/A
42.07
100.28
MVSFormer++
15.34
64.99
79.29
80.42
304.28
Table 3: Runtime (s) of different methods on Island across all subsets with varying numbers of input images (2, 10, 20, 50, and 100). “N/A” indicates no result is available.
Methods
2
10
20
50
100
DUSt3R
3.39
42.93
142.54
MASt3R
13.16
39.17
88.61
1086.80
VGGT
1.63
5.42
10.56
29.67
63.06
Fast3R
0.71
0.83
1.72
5.47
15.59
MVSAnywhere
N/A
N/A
N/A
57.47
101.51
MVSFormer++
14.57
34.47
55.53
121.09
268.59
Table 4: Runtime (s) of different methods on Urban area across all subsets with varying numbers of input images (2, 10, 20, 50, and 100). “N/A” indicates no result is available.
Images
Methods
RMS (GSD)
MD (GSD)
Coverage (%)
2
DUSt3R
2.88
2.21
84.89
MASt3R
4.53
3.26
71.85
Fast3R
8.76
6.85
38.26
VGGT
3.71
2.58
77.77
10
DUSt3R
9.33
7.19
38.85
Fast3R
12.73
7.72
46.02
Table 5: Quantitative results of different methods on the Island scene under various numbers of input images. GSD = 0.267m/pixel.
Images
Methods
RMS (GSD)
MD (GSD)
Coverage (%)
2
DUSt3R
5.50
4.11
55.26
MASt3R
3.87
2.62
78.53
Fast3R
6.74
5.14
45.91
VGGT
5.60
3.83
65.06
10
DUSt3R
3.69
2.55
75.69
Fast3R
5.35
3.83
62.09
Table 6: Quantitative results of different methods on the Urban scene under various numbers of input images. GSD = 0.282 m/pixel
Figure 4: Comparison of reconstruction accuracy on the Island scene using DUSt3R and VGGT with 2 and 20 input images. The colored maps show point-wise distance errors with respect to the LiDAR ground truth. Blue regions indicate small deviations (higher accuracy), whereas red regions denote larger errors. The color scale represents the error magnitude in meters.
Figure 5: Comparison of reconstruction accuracy on the Urban scene using DUSt3R and VGGT with 2 and 20 input images. The colored maps show point-wise distance errors with respect to the Pix4dmapper-generated ground truth. Blue regions indicate small deviations (higher accuracy), whereas red regions denote larger errors. The color scale represents the error magnitude in meters.
Illusory matches between distinct yet visually similar 3D surfaces--doppelgangers--remain a fundamental obstacle for large-scale, in-the-wild 3D reconstruction and visual localization. Prior work mitigates this issue with pairwise classifiers, but this design limits multi-view contextual reasoning and incurs O(n^2) inference complexity for downstream structure-from-motion (SfM). We present MVDG, a scalable multi-view disambiguation framework built on the 3D foundation model VGGT, which jointly reasons over an arbitrary number of multiview images. By incorporating 3D-aware multi-view features, our method reduces dependence on pairwise comparisons by encoding and decoding views in a single pass. We further observe that direct multi-view fine-tuning of VGGT can be unstable under noisy supervision; motivated by label ambiguity in Doppelgangers, we construct a pseudo-pairwise training set from AerialMegaDepth and show that fine-tuning on sampled subsets yields stable optimization and strong generalization to held-out scenes. Finally, because full SfM evaluation (even with faster pipelines such as GLOMAP) remains expensive, we process a pseudo-pairwise dataset for efficient validation; we derive a predictive relationship between regular SfM metrics and the classification accuracy on this pseudo-pairwise test. Experiments show that our method achieves comparable pairwise accuracy while improving both SfM accuracy and inference speed over baselines.
Hanyuan Xiao, Gonglin Chen, Haolin Xiong +3
University of Southern California · Institute for Creative Technologies
Structure from Motion (SfM) is essential for multi-view 3D reconstruction, however, its accuracy heavily relies on the accuracy of image matching. While the recent correspondence matching method, MASt3R, enables robust matching even under challenging conditions, it tends to generate incorrect correspondences for non-overlapping image pairs. Consequently, existing SfM methods using MASt3R, such as MASt3R-SfM, suffer from significant degradation in pose estimation accuracy as they incorporate these unreliable matches directly into optimization. To address this issue, we propose G-MASt3R-SfM, a novel SfM pipeline that enhances robustness through two key modules. First, the Graph-based View Pruning (GVP) module constructs a scene graph from matching confidence and geometrically prunes outlier views. Second, the Multi-Stage Optimization (MSO) module progressively refines camera parameters by expanding the optimization scope from local consistency to the global consistency. Experiments on the ETH3D dataset demonstrate that our method achieves state-of-the-art accuracy in both camera pose estimation and 3D reconstruction, effectively suppressing noise caused by outliers.
Toshiki Watanabe, Shintaro Ito, Natsuki Takama +2
Graduate School of Information Sciences, Tohoku University, Japan
Establishing consistent correspondences across images is essential for 3D vision tasks such as structure-from-motion (SfM), yet most existing matchers operate in a pairwise manner, often producing fragmented and geometrically inconsistent tracks when their predictions are chained across views. We propose MV-RoMa, a multi-view dense matching model that jointly estimates dense correspondences from a source image to multiple co-visible targets. Specifically, we design an efficient model architecture which avoids high computational cost of full cross-attention for multi-view feature interaction: (i) multi-view encoder that leverages pair-wise matching results as a geometric prior, and (ii) multi-view matching refiner that refines correspondences using pixel-wise attention. Additionally, we propose a post-processing strategy that integrates our model's consistent multi-view correspondences as high-quality tracks for SfM. Across diverse and challenging benchmarks, MV-RoMa produces more reliable correspondences and substantially denser, more accurate 3D reconstructions than existing sparse and dense matching methods. Project page: https://icetea-cv.github.io/mv-roma/.