V3DSAM: Semantic-Geometric Loop Verification and Geometry-Anchored Persistent Instance Mapping for Feed-Forward Dense SLAM
Organizations: Information Engineering University
Abstract
To address unreliable cross-view association in long-sequence feed-forward dense SLAM, this letter presents V3DSAM, an open-vocabulary semantic dense SLAM system that establishes physical correspondences at both the scene and object levels. The system combines feed-forward geometric estimates from VGGT-SLAM 2.0 with open-vocabulary semantic observations from SAM3. We introduce semantic-geometric dual-scale loop verification, which identifies false loop closures using pixel-level bidirectional reprojection and object-level 3D spatial correspondences, without accessing backbone-internal features. We further propose geometry-anchored persistent instance mapping, which defines physical instances through class-constrained 3D connectivity in an optimized global coordinate system and assigns global IDs without inter-frame identity propagation. The proposed verifier rejects all four false loop candidates in our TUM RGB-D evaluation, while long-sequence ScanNet experiments show improved instance identity consistency. Thus, V3DSAM reduces dependence on internal backbone features and temporal ID propagation for robust long-term cross-view association.
Figures & tables
| Verification signal | FP | Precision | F1 |
| SALAD retrieval | 4 | 0.733 | 0.846 |
| VGGT | 0 | 1.000 | 1.000 |
| Semantic only (Category Jaccard) | 2 | 0.846 | 0.917 |
| Geometry only (Depth consistency) | 2 | 0.846 | 0.917 |
| Semantic-Geometric 3D (Ours) | 0 | 1.000 | 1.000 |
| Setting | Method | 360 | desk | desk2 | plant | room | rpy | teddy | xyz |
| Calibrated | ORB-SLAM3 | - | 0.017 | 0.210 | 0.034 | - | - | - | 0.009 |
| DeepV2D | 0.243 | 0.166 | 0.379 | 0.203 | 0.246 | 0.105 | 0.316 | 0.064 | |
| DeepFactors | 0.159 | 0.170 | 0.253 | 0.305 | 0.364 | 0.043 | 0.601 | 0.035 | |
| DPV-SLAM | 0.112 | 0.018 | 0.029 | 0.021 | 0.330 | 0.030 | 0.084 | 0.010 | |
| DPV-SLAM++ | 0.132 | 0.018 | 0.029 | 0.022 | 0.096 | 0.032 | 0.098 | 0.010 | |
| GO-SLAM | 0.089 | 0.016 | 0.028 | 0.026 | 0.052 | 0.019 | 0.048 | 0.010 |
| Scene | Frames | Loops | False | ATE b | ATE a |
| scene0059_00 | 1787 | 7 | 0 | 0.091 | 0.091 |
| scene0000_00 | 5578 | 12 | 0 | 0.140 | 0.140 |
| scene0106_00 | 2324 | 4 | 0 | 0.121 | 0.121 |
| scene0207_00 | 1988 | 8 | 2 | 0.109 | 0.089 |
| Method | IDC | ReID-R | Inst. Coverage | MOTA | 4D-PQ | IDS |
| SAM3 Tracking | 0.146 | 0.000 | 0.223 | -0.184 | 0.118 | 285 |
| CLIP Association | 0.189 | 0.058 | 0.594 | -6.671 | 0.244 | 3263 |
| OVO (Vp+Vd) | 0.439 | 0.183 | 0.975 | -13.019 | 0.026 | 2994 |
| V3DSAM (Ours) | 0.565 | 0.192 | 0.736 | -0.286 | 0.110 | 873 |
| OVO (GTp+GTd) | 0.631 | 0.317 | 0.918 | -4.603 | 0.046 | 1488 |
| V3DSAM (GTp+GTd) | 0.745 | 0.481 | 0.903 | -2.805 | 0.054 | 901 |
Simultaneous Localization and Mapping
Simultaneous localization and mapping estimates a robot’s motion while building a map of an unknown environment.
See all 210 papers