Abstract
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.
The modules highlighted in green (GVP and MSO) represent our novel contributions to the standard MASt3R-SfM pipeline. Utilizing correspondences and confidence maps from MASt3R, the GVP module constructs a scene graph to filter out unreliable views, followed by the MSO module which refines camera parameters..
Qualitative comparison of 3D reconstruction results. The red boxes highlight that the proposed method effectively eliminates artifacts observed in MASt3R-SfM (top and middle rows) while successfully reconstructing texture-less regions where COLMAP fails (bottom row).
Experimental Results
Quantitative results of camera pose estimation on the ETH3D dataset. The best results are highlighted in bold.
| Method | RRE ↓ [deg.] |
RTE ↓ [deg.] |
AUC@5 ↑ [%] |
SfM rate [%] |
|---|---|---|---|---|
| COLMAP | 0.655 | 2.645 | 90.7 | 87 |
| DFSfM | 2.298 | 3.711 | 68.0 | 85 |
| VGGSfM | 22.439 | 17.220 | 52.7 | 98 |
| VGGT | 2.485 | 8.806 | 35.4 | 100 |
| MASt3R-SfM | 2.572 | 3.343 | 75.7 | 100 |
| G-MASt3R-SfM (Ours) | 0.474 | 0.978 | 93.9 | 97 |
BibTeX
@article{Watanabe-ICIP-2026,
title={G-MASt3R-SfM: Graph-based View Pruning and Multi-stage Optimization for Robust SfM},
author={Toshiki Watanabe and Shintaro Ito and Natsuki Takama and Koichi Ito and Takafumi Aoki},
journal={IEEE International Conference on Image Processing},
year={2026},
url={https://gsisaoki.github.io/G-MASt3R-SfM/}
}