Dense metric mapping from a single camera and IMU remains challenging: visual-inertial odometry (VIO) provides metric motion but sparse geometry, whereas feed-forward depth models predict dense structure without ensuring metric scale. We present Accord, which constructs an accurate online dense metric map from any camera with an IMU. A VIO filter publishes metric poses, a depth foundation model conditioned on them predicts depth for short keyframe windows, and a factor graph fuses the resulting submaps. Recent work has explored several approaches to this problem. What is essential is the multi-view agreement (Accord), which is considered in SLAM as a source of geometric self-consistency. Accord anchors each keyframe window to metric 3D points triangulated from the filter poses, then exploits bidirectional dense depth agreement between co-visible images to estimate and correct per-image scale errors while preserving the metric scale established by the anchors. Context keyframes further improve depth prediction, and the resulting submaps are fused in a factor graph. We evaluate a single mapping configuration across five public camera-IMU setups and a new Phone-rig/GoPro dataset. With nothing tuned per dataset/setup and every map aligned only by Accord's own trajectory, Accord achieves consistently best/second-best mean F-score at 5 cm on every platform. Accord is open-source along with the phone-rig dataset for the benefit of the research community
Figures & tables
Figure 2: Accord Architecture: three processes . Gray: the base system [ 19 ] ; orange: Accord (ours).
Setup
Platform, scene
Camera
Seq.
Time, path
EuRoC
drone, rooms
gray
6
11 min, 0.4 km
TUM-VI
hand-held, lab
gray fisheye
6
14 min, 0.7 km
ORI
hand-held, buildings
color fisheye
5
11 min, 0.5 km
Hilti-Oxford
survey rig, heritage
gray fisheye
5+5
26 min, 1.0 km
Oxford Spires
hand-held, outdoors
color fisheye
3
28 min, 1.7 km
Phone-rig (ours)
two smartphones, campus
color
11+10
59 min, 2.5 km
Table 1: Platform, camera, sequences, total time and path
Dense map: F-score at 5 cm ↑
Trajectory error (cm) ↓
EuRoC
ORI
Hilti-Oxford
Oxford Spires
EuRoC
TUM-VI
Hilti-Oxford
Oxford Spires
System
drone, rooms
hand-held, buildings
survey rig
outdoors
drone, rooms
hand-held
survey rig
outdoors
ORB-SLAM3 [ 3 ]
–
–
–
–
3.2
2.0
68.2
–
OKVIS2-X [ 1 ]
–
–
–
–
6.0
1.4
17.1
–
MASt3R-Fusion [ 41 ]
0.41
0.53
0.13
×
7.0
3.0
83.7
–
VGGT-SLAM 2.0 [ 17 ] †
0.49
0.18
0.04
0.05
7.4
25.0
438.3
382.5
Table 2: The public benchmarks. Means over each dataset’s sequences; bold : best, underlined : second best. Left: the dense map’s F-score at 5 cm. Right: the trajectory’s error in cm. × : every sequence failed (no output, or a trajectory more than 10 m drifted off)
Figure 3: Every scanned sequence. The map’s F-score at 5 cm per sequence, placed by each system’s own trajectory; the means are Tab. 2 ’s. A × at 0: no map, or the map of a trajectory that failed (more than 10 m off). † , ‡ as in Tab. 2 .
Figure 4: Maps against the laser scan. A top-view horizontal slice of each pipeline’s map (light blue: scanned but not mapped). Every map point is colored by its distance to the scan. F is the sequence’s F-score at 5 cm
Figure 5: Accord against FoundationVIO outdoors (Oxford Spires CC05).
Figure 6: The phone rig dataset. The 11 campus walks span indoor and outdoor environments, including two walks at night. Bottom right: the rig. The Note 9 (right) and the LiDAR iPhone that runs RTAB-Map [ 10 ] (left), whose result is the reference. The phones are carried by hand on this frame and are not calibrated ( Appendix G ).
Note 9, 10 walks
iPhone, 11 walks
System
F@5 ↑
F@10 ↑
ATE (m) ↓
F@5 ↑
F@10 ↑
ATE (m) ↓
ORB-SLAM3 [ 3 ]
–
–
0.71
–
–
0.88
OKVIS2-X [ 1 ]
–
–
–
–
–
0.34
MASt3R-Fusion [ 41 ]
0.04
0.08
1.88
×
×
–
VGGT-SLAM 2.0 [ 17 ] †
0.07
0.14
4.76
0.09
0.16
2.72
ScaRF-SLAM’s mapper [ 38 ] ‡
0.14
0.26
–
0.10
0.18
–
Table 3: The phone benchmark. Means over each phone’s recordings: map F-score at 5 and 10 cm, ATE(m).
Figure 7: A GoPro through a shopping mall.
EuRoC rooms
Hilti-Oxford
DAVIO [ 19 ]
0.381
–
our mapper, no anchor or context
0.488
0.120
+ the anchor
0.547
0.147
+ context keyframes (Accord)
0.570
0.164
Table 4: Ablation : the parts of our mapper added one at a time, with DA3-Giant showing Mean F-score at 5 cm of the map.
Depth model
EuRoC Rooms, F
ORI, F
GPU (GB)
DA3-Giant
0.570
0.782
8.9
DA3-Large
0.521
0.698
5.1
DA3-Base
0.499
0.677
2.5
Table 5: Three depth models. F-score at 5 cm.
System
Speed (FPS)
GPU (GB)
RAM (GB)
Accord
18
11.1
2.4
FoundationVIO
15
13.2
7.1
MASt3R-Fusion
12
4.9
6.8
VGGT-SLAM 2.0
23
6.8
9.9
ScaRF-SLAM’s mapper
15
9.4
5.6
Table 6: Speed and memory on EuRoC, one RTX 4090.
Appendix figures & tables20 assets
Supplementary material from the paper’s appendix.
Appendix
Figure 8: The two-way measurement. (a) A scale error: image i puts a surface point too far; seen from j , the point lies behind the surface that j ’s own depth dj reads. Carried the other way the sign flips. (b) An occlusion: a point one camera sees is hidden from the other, which reads the nearer occluder. The sign is the same both ways, so half the difference of the two directions keeps the scale error and removes what the occlusions share.
Figure 9: Before and after the correction (ORI R02, a live run): a horizontal slab of one floor from above, placed by the system’s own trajectory. Every map point is colored by its distance to the scan, drawn under the map. Below, the framed part enlarged. As built (left), images that disagree in scale by a few percent smear a wall into layers. After the correction they agree.
Setting
Value
Windows
keyframes per window
5
keyframes two windows share
2
seconds between keyframes
0.5
seconds a window may wait for the worker
2
Context keyframes
Appendix
Table 7: The settings of the system , the same on every platform.
EuRoC rooms
ORI
System
traj.
rigid
sim.
traj.
rigid
sim.
Accord, DA3-Giant
0.570
0.601
0.632
0.782
0.788
0.852
Accord, DA3-Large
0.521
0.561
0.588
0.698
0.711
0.798
Accord, DA3-Base
0.499
0.539
0.562
0.677
0.693
0.751
MASt3R-Fusion [ 41 ]
0.411
0.434
0.442
0.525
0.533
0.556
VGGT-SLAM 2.0 [ 17 ] †
0.490
0.539
0.547
0.184
0.198
0.202
Appendix
Table 8: Dense maps under three placements , mean F-score at 5 cm on the 6 EuRoC rooms and the 5 ORI sequences: placed by the system’s own trajectory, after a rigid registration to the scan, and after a similarity. † no metric scale: given the reference’s. ‡ with its own depth model, on our poses, first placement only. –: not run.
Accord
MASt3R-Fusion
VGGT-SLAM 2.0 †
ScaRF ‡
Seq.
traj.
rigid
sim.
traj.
rigid
sim.
traj.
rigid
sim.
traj.
V1_01
0.536
0.569
0.649
0.457
0.458
0.458
0.580
0.628
0.627
0.251
V1_02
0.728
0.734
0.744
0.542
0.548
0.550
0.476
0.505
0.522
0.133
V1_03
0.518
0.550
0.591
0.476
0.483
0.480
0.456
0.507
0.512
0.130
V2_01
0.513
0.588
0.635
0.216
0.327
0.341
0.456
0.575
0.582
0.328
V2_02
0.614
0.626
0.627
0.362
0.363
0.378
0.476
0.522
0.528
0.179
Appendix
Table 9: The maps of the systems, per sequence : F-score at 5 cm placed by the trajectory, after a rigid registration and after a similarity. Accord: live, with DA3-Giant. † , ‡ as in Tab. 8 .
V1_01
V1_02
V1_03
V2_01
V2_02
V2_03
mean
ORB-SLAM3 [ 3 ]
0.020
0.025
0.054
0.048
0.028
0.021
0.032
OKVIS2 [ 1 ] , final
0.034
0.021
0.040
0.131
0.055
0.077
0.060
MASt3R-Fusion, real-time
0.146
0.115
0.093
0.268
0.129
0.312
0.177
MASt3R-Fusion, global ‡
0.064
0.028
0.039
0.195
0.042
0.052
0.070
VGGT-SLAM 2.0 †
0.058
0.080
0.077
0.079
0.066
0.083
0.074
Accord, the filter
0.040
0.054
0.057
0.066
0.053
0.187
0.076
Appendix
Table 10: Trajectories on the EuRoC rooms : ATE in meters after a rigid alignment, the median of the runs of the other systems and the mean of Accord’s. † after a similarity, on its keyframes. ‡ on its keyframes.
F-score at 5 cm
ATE (m)
Windows
Model
Seq.
runs
built
corr.
rigid
sim.
filter
final
mapped
not
loops
Giant
V1_01
3
0.588
0.536
0.569
0.649
0.040
0.041
88.3
9.3
3.3
Giant
V1_02
3
0.658
0.728
0.734
0.744
0.054
0.044
53.0
0.3
7.0
Giant
V1_03
3
0.486
0.518
0.550
0.591
0.057
0.054
66.0
0.0
6.0
Giant
V2_01
3
0.503
0.513
0.588
0.635
0.066
0.066
70.0
2.7
0.0
Giant
V2_02
3
0.510
0.614
0.626
0.627
0.053
0.051
73.3
4.3
9.3
Appendix
Table 11: Accord live, per sequence and depth model : F-score at 5 cm of the map as built, corrected, and after a rigid and a similarity registration of the corrected map; ATE of the filter and of the trajectory corrected by the final map; windows mapped and not mapped (dropped while the worker was busy, or taken with the camera nearly at rest); loops closed. Means over the runs.
Figure 10: Precision and recall at 5 cm , the map placed by the system’s own trajectory: every sequence (small marks) and the mean (large marks); the curves are lines of equal F-score. † , ‡ as in Tab. 8 .
Sequence
Site
Reference
Scan
development
exp14 Basement 2
building
dense
yes
exp11 Lower Gallery
building
points
yes
exp15 Attic to Upper Gallery
building
points
yes
exp05 Construction Upper Level 2
site
points
no
exp07 Long Corridor
site
points
none
exp02 Construction Multilevel
site
points
–
Appendix
Table 12: The Hilti-Oxford sequences we use. Reference: dense poses and control points, or control points only. Scan: whether the map is scored (yes), or why not: the site’s scan does not match the sequence (no), there is no scan (none), or the sequence was run with the filter alone and has no map (–).
Noise
RMSE
the dataset’s first configuration
0.627
densities as measured, × 1 / 1.5 / 2 / 3
0.167 / 0.201 / 0.296 / 0.415
and walks at their bounds, × 1 / 3 / 10
0.154 / 0.159 / 0.173
Appendix
Table 13: The IMU’s noise on Hilti-Oxford , the filter alone: mean RMSE (m) at the control points of the five development sequences.
development
evaluation
System
exp14
exp05
exp07
exp11
exp15
exp18
exp16
exp04
exp01
exp21
ORB-SLAM3 [ 3 ]
fail
0.170 / 1.7
fail
0.291 / 3.0
0.234 / 2.2
0.127
0.157 / 3.3
fail
0.948 / 0.0
1.495 / 0.0
OKVIS2 [ 1 ] , final
0.054 / 27.5
0.162 / 5.0
0.589 / 0.0
0.147 / 6.0
0.224 / 0.0
0.205 / 20.0
0.229 / 3.3
0.108 / 22.9
0.104 / 15.4
0.210 / 0.0
MASt3R-Fusion [ 41 ]
0.178 / 0.0
0.368 / 0.0
0.808 / 0.0
0.120 / 8.0
0.328 / 0.0
0.139
div.
0.458 / 1.4
1.578 / 0.0
1.174 / 0.0
VGGT-SLAM 2.0 [ 17 ] †
1.009 / 0.0
4.737 / 0.0
div.
2.592 / 0.0
6.631 / 0.0
6.464
4.647 / 0.0
4.689 / 0.0
3.690 / 0.0
2.424 / 0.0
OKVIS2, live
0.093 / 20.0
0.173 / 1.7
0.785 / 0.0
0.230 / 4.0
0.305 / 0.0
0.230 / 20.0
0.322 / 0.0
0.195 / 4.3
0.160 / 10.0
0.464 / 0.0
Appendix
Table 14: Trajectories on Hilti-Oxford : the error in meters (ATE where the reference is dense, exp14, exp18 and exp16; the RMSE at the control points elsewhere) and the dataset’s score. fail: no run initialized; div.: more than 10 m off (failed). † given the reference’s scale.
development
evaluation
exp14
exp11
exp15
exp18
exp16
exp21
System
traj.
rigid
traj.
rigid
traj.
rigid
traj.
rigid
traj.
rigid
traj.
rigid
MASt3R-Fusion [ 41 ]
0.313
0.409
0.180
0.281
0.183
0.199
0.288
0.299
div.
div.
0.093
0.186
VGGT-SLAM 2.0 [ 17 ] †
0.076
0.088
0.064
0.064
0.031
0.034
0.032
0.033
0.028
0.028
0.057
0.064
ScaRF-SLAM’s mapper [ 38 ] ‡
0.209
0.242
0.153
0.157
0.073
0.129
0.120
0.129
0.107
0.112
0.053
0.138
Accord, DA3-Base
0.384
0.398
0.205
0.277
0.174
0.232
0.176
0.213
0.179
0.181
0.066
0.147
Appendix
Table 15: Maps on Hilti-Oxford , the sequences whose scan matches them: F-score at 5 cm placed by the system’s own trajectory, and after a rigid registration to the scan. div.: the trajectory more than 10 m off (failed). † , ‡ as in Tab. 8 .
System
room1
room2
room3
room4
room5
room6
mean
ORB-SLAM3 [ 3 ]
0.021
0.018
0.040
0.018
0.013
0.007
0.020
OKVIS2 [ 1 ] , final
0.012
0.014
0.012
0.013
0.013
0.021
0.014
OKVIS2, live
0.024
0.019
0.057
0.023
0.281
0.033
0.073
MASt3R-Fusion, real-time
0.168
0.519
0.513
0.139
0.196
0.313
0.308
MASt3R-Fusion, global §
0.020
0.037
0.041
0.022
0.019
0.041
0.030
VGGT-SLAM 2.0 †
0.247
0.268
0.467
0.126
0.209
0.182
0.250
Appendix
Table 16: Trajectories on TUM-VI : ATE in meters after a rigid alignment, the median of the runs of the other systems and the mean of Accord’s. fail: no pose. † after a similarity, on its keyframes. § on its keyframes.
Accord: map
FoundationVIO: map
Trajectory (m)
Seq.
built
corrected
P
R
F
P
R
Accord
filter
FVIO
CC02
0.131
0.137
0.16
0.12
0.082
0.13
0.06
1.09
0.63
1.13
CC03
0.242
0.265
0.25
0.29
0.191
0.28
0.15
0.30
0.39
0.16
CC05
0.185
0.187
0.21
0.17
0.074
0.11
0.06
0.80
0.82
1.86
Mean
0.186
0.196
0.21
0.19
0.116
0.17
0.09
0.73
0.61
1.05
Appendix
Table 17: Oxford Spires, against FoundationVIO : F-score at 5 cm of the map placed by its own trajectory, with its precision (P) and recall (R); Accord as built and corrected (means of three runs), FoundationVIO’s one run. Trajectory error (m) after a rigid fit: Accord’s corrected by the map and as the filter published it, FoundationVIO’s as it outputs it.
Map: F-score at 5 cm ↑
Trajectory (m) ↓
CC02
CC03
CC05
mean
CC02
CC03
CC05
mean
MASt3R-Fusion
–
–
–
0.000 3
failed
failed
failed
– 3
VGGT-SLAM 2.0 †
0.042
0.096
–
0.046 1
7.12
0.53
failed
3.82 1
ScaRF mapper ‡
0.044
0.063
0.041
0.049
–
–
–
–
FoundationVIO
0.082
0.191
0.074
0.116
1.13
0.16
1.86
1.05
DAVIO, as released
–
–
–
0.000 3
0.87
0.52
failed
0.70 1
Appendix
Table 18: Every dense pipeline on Oxford Spires : the F-score at 5 cm of the map placed by its own trajectory, and the trajectory’s error (m). Accord and DAVIO: means of three runs; the others: one run. –: no map, or the map of a failed trajectory; failed: a trajectory more than 10 m off. In the means, a map not made or of a failed trajectory counts 0 ( n : n such), and a failed trajectory is left out ( n : n left out). † given the reference’s scale. ‡ on our poses, so no trajectory of its own.
F-score at 5 cm (%), its protocol
APE (m)
FoundationVIO
Accord
FoundationVIO
Seq.
its paper
our run: tool
our run: its rule
tool
its rule
its paper
our run
R01
76.2
63.1
67.9
72.8
77.4
0.188
0.272
R02
74.4
61.0
65.4
66.1
71.9
0.070
0.093
R03
75.7
61.8
65.4
72.9
77.2
0.064
0.128
R04
72.6
53.7
57.1
70.5
74.3
0.127
0.255
Appendix
Table 19: FoundationVIO against its paper on ORI. F-score at 5 cm (%) under the protocol of its paper (ScaRF-SLAM’s compare_pts.py ): as the tool is released, and with the rule its paper states for recall. Accord: the corrected map. APE (m) after a rigid fit: as its paper reports it, and our run scored by its own evaluation script.
Walk
Scene
Time
Path (m)
Closures
Note 9 (s)
iPhone (s)
ARKit (m)
VIO01
indoor: corridors of a building
3:36
128
28
221
120
0.15
VIO02
indoor: corridors and a stairwell over several floors
6:32
249
44
394
130
0.17
VIO03
outdoor: paved courtyard with bicycle racks
3:58
180
10
–
238
0.23
VIO04
outdoor: courtyard with planters and steps
5:13
223
3
316
120
0.14
VIO05
outdoor: road around a block, parked cars
4:45
283
41
286
238
0.44
VIO06
outdoor: park with playground equipment
6:30
294
58
396
120
0.14
Appendix
Table 20: The walks of the phone rig dataset. Time and path are the reference’s; closures: RTAB-Map’s loop closures; each phone’s span is its recording’s (the Note 9 has none of VIO03, whose camera stopped after a few seconds; the iPhone stopped early on most walks); ARKit: the iPhone’s own tracking against the reference after a rigid fit, over the iPhone’s part of the walk. Below the rule, the walks added after the first 9 had been benchmarked.
Recording
MASt3R-F.
VGGT-SLAM †
ScaRF ‡
FVIO
DAVIO
Accord
Note 9
VIO01
0.19
0.06
0.21
0.17
0.26
0.41
VIO02
0.00
0.02
0.00
0.11
0.00
0.00
VIO04
0.00
0.15
0.17
0.08
0.14
0.26
VIO05
0.08
0.06
0.08
0.07
0.08
0.11
VIO06
0.00
0.15
0.17
0.19
0.26
0.30
Appendix
Table 21: The phone rig, maps : per recording, the F-score at 5 cm against RTAB-Map’s map, placed by the system’s own trajectory, points beyond 3 m of the reference track left out (their share, in %, below each phone). 0.00: no map, or the map of a failed trajectory. Accord and DAVIO: the mean of three runs. Bold: the best of the recording. † given the reference’s scale; ‡ on our poses.
Figure 11: The maps of four pipelines on the EuRoC rooms : a horizontal slab 1.0 to 1.4 m above the floor, seen from above. Each map is placed by its own trajectory and each of its points is colored by its distance to the scan, which is drawn under the map and shows where nothing was mapped. F: the F-score at 5 cm of the whole map ( Tab. 9 ). Accord: the live run of the median score. † , ‡ as in Tab. 8 .
Figure 12: The maps of four pipelines on ORI , drawn as in Fig. 11 .
A camera and an IMU are the minimal sensor setup for metric localization and dense mapping, yet classical visual--inertial filters must wait for parallax before they start and then retain only sparse landmarks. Feed-forward geometry models, in contrast, predict dense structure from a few images but provide neither metric scale nor gravity. We present DAVIO, which uses a single multi-view depth model, Depth Anything~3, for both start-up and mapping. At start-up, a five-image window and preintegrated IMU measurements form a feature-free linear system. Its robust, conditioning-checked solution bootstraps a VIO filter through buffered replay. During tracking, the filter's metric poses condition the depth model. Residual scale is corrected only along viewing rays, which preserves the metric camera baselines, and a gravity-preserving submap graph with drift-gated revisits refines the map. On EuRoC, DAVIO starts markedly earlier, reduces the localization error, and maps more accurately than SOTA feed-forward mappers given identical poses. On building-scale ORI sequences, DAVIO is on bar or better than SOTA mappers on the same odometry, and degrades far less when GT poses are replaced by real odometry. We release the code of DAVIO, a real-time dense metric SLAM system, to the community.
Jaafar Mahmoud, Arthur Movsesyan, Mikhail Iumanov +1
Biomechatronics and Energy-Efficient Robotics (BE2R) Lab, ITMO University, Saint Petersburg, Russia.
Monocular foundation models provide dense geometry but usually lack a stable metric scale. This paper presents VIDAR, a visual-inertial dense reconstruction framework that couples SVO+IMU odometry with Depth Anything 3. VIDAR uses the visual-inertial front end as a metric anchor: it provides camera poses, scale, and a consistent world frame for aligning dense foundation-model predictions across time. The foundation model then contributes detailed local geometry that is fused into a global reconstruction. We study both pose-conditioned DA3 and a decoupled alignment strategy. On EuRoC, pose injection reduces scale error to about 1% and reaches 0.463 mean [email protected]; the decoupled hybrid improves this to 0.676 without ground-truth poses. Results on EuRoC and TUM RGB-D show that VIDAR is a practical route to metric dense monocular reconstruction.
Diyari Mohammed Salih, Lingxiang Hu, Naima AitOufroukh-Mammar +1
IBISC Laboratory, Université d’Évry Paris-Saclay, Université Paris-Saclay, Évry-Courcouronnes, France.
We present a stereo visual odometry (SVO) framework that integrates geometric feed-forward models (GFMs) for dense depth prediction with factor-graph-based visual odometry to achieve accurate localisation and dense metric mapping. Our key idea is to exploit sparse but reliable metric geometry estimated from stereo and temporal multi-view observations to calibrate dense GFM predictions, while retaining the robustness and efficiency of model-based SVO. Rather than directly conditioning the GFM on potentially noisy stereo depth, optimised 3D landmarks are used to jointly estimate a scale-shift transformation over all keyframes within each local submap, yielding metrically consistent dense depth predictions. To balance accuracy and computational efficiency, we introduce a hierarchical factor-graph architecture. At the local level, sliding-window optimisation estimates high-frequency camera poses and multi-view-refined 3D landmarks. Dense predictions are generated only for selected keyframes and registered within each submap using SE(3) Generalised ICP. Consecutive submaps are further aligned in Sim(3) using their shared geometry, allowing residual scale inconsistencies to be corrected. These intra- and inter-submap constraints are incorporated into a global factor graph to produce a consistent trajectory and dense metric reconstruction. Extensive experiments on public and in-house datasets demonstrate that the proposed framework achieves competitive odometry accuracy against geometric and learning-based baselines while providing accurate dense metric reconstruction and improved robustness to failures in learned depth prediction.
Fahira Afzal Maken, Sagun Man Singh Shrestha, Edward Vanderfeen +4
CSIRO Robotics, Technology, CSIRO, Brisbane, Australia. · School of Electrical Engineering and Robotics, Queensland University of Technology (QUT), Brisbane, Australia.