Geometric Foundation Models (GFMs) have recently advanced monocular SLAM by providing robust, calibration-free 3D priors. However, deploying these models on dense video streams introduces significant computational redundancy. Current GFM-based SLAM systems typically rely on post hoc keyframe selection. Because of this, they must perform expensive dense geometric decoding simply to determine whether a frame contains novel geometry, resulting in late rejection and wasted computation. To mitigate this inefficiency, we propose LeanGate, a lightweight feed-forward frame-gating network. LeanGate predicts a geometric utility score to assess a frame's mapping value prior to the heavy GFM feature extraction and matching stages. As a predictive plug-and-play module, our approach bypasses over 90% of redundant frames. Evaluations on standard SLAM benchmarks demonstrate that LeanGate reduces tracking FLOPs by more than 85% and achieves a 5x end-to-end throughput speedup. Furthermore, it maintains the tracking and mapping accuracy of dense baselines. Project page: https://lean-gate.github.io/
Figures & tables
Method
desk
360
dsk2
rpy
xyz
flor
plnt
room
tdy
Avg
ORB3 [ 57 ]
–
1.7
21.0
–
0.9
–
3.4
–
–
–
DPV [ 58 ]
11.2
1.8
2.9
3.0
1.0
5.7
2.1
33.0
8.4
7.6
DROID [ 15 ]
11.1
1.8
4.2
2.6
1.2
2.1
1.6
4.9
4.8
3.8
MASt3R-SLAM [ 20 ]
Full(15FPS)
1.6
4.9
2.3
2.7
0.9
2.5
2.0
6.1
4.1
3.0
Keyframe(0.05FPS)
1.5
4.9
2.6
2.7
0.8
2.6
1.9
6.6
3.7
3.0
TABLE I : Absolute Trajectory Error (ATE in cm) on TUM-RGBD sequence fr1 comparing full-frame tracking at 15 FPS with keyframe-only tracking; lower is better.
Fig. 2 : Sim(3) -aligned trajectory comparison on TUM-RGBD fr1-teddy , contrasting full-frame (15 FPS) tracking with keyframe-only tracking.
Fig. 3 : ATE and runtime under naive stride policies across RGB benchmarks. The shaded region marks the stride regime where TUM starts to lose scenes; red crosses denote policies evaluated on fewer scenes than the full set.
Fig. 4 : Averaged per-scene time breakdown profiled on TUM-RGBD without frame/model loading time. Tracker (encoder/decoder) takes most of the computation time.
Fig. 5 : System Overview for LeanGate. Left: score-only distillation. Right: MASt3R-SLAM slimmed by LeanGate. The most recent keyframe (in brown color) is fed to the LeanGate to calculate the score for the newly arriving frame (in grey).
Dataset
Method
Downsample
Comp ↓ ( Δ%↑ )
Chamfer ↓ ( Δ%↑ )
F@2cm ↑ ( Δ%↑ )
F@5cm ↑ ( Δ%↑ )
TUM RGB-D
DROID-SLAM [ 15 ]
1×
0.631
0.355
0.076
0.184
MASt3R-SLAM [ 20 ]
All
1×
0.107
0.143
0.204
0.425
Stride 2
2×
0.129 ( −20.6 )
0.145 ( −1.4 )
0.212 (+3.9)
0.433 (+1.9)
Stride 15
15×
0.545 ( −409.3 )
0.393 ( −174.8 )
0.188 ( −7.8 )
0.368 ( −13.4 )
LeanGate
16×
0.160 ( −49.5 )
0.149 ( −4.2 )
0.202 ( −1.0 )
0.422 ( −0.7 )
TABLE II: Reconstruction quality under different downsampling strategies on TUM RGB-D, EuRoC MAV, and 7-Scenes. We compare LeanGate against uniform striding; metrics include Completion (Comp), Chamfer distance, and F-score at 2 cm and 5 cm (higher is better for F-score, lower is better for distances). Δ % denotes change relative to the 1× baseline; LeanGate preserves quality with 16× – 32× fewer frames, approaching the 2× stride baseline.
Dataset
Model
Downsample
ATE [cm] ↓
Time [s] ↓
Calculations [TFLOPs] ↓
Frame Select
SLAM
Total
TUM RGB-D
DPV-SLAM [ 58 ]
1×
7.6
43.79
—
—
—
DROID-SLAM [ 15 ]
1×
3.8
41.78/39.56
—
—
—
MASt3R-SLAM [ 20 ]
2×
3.00
74.95
—
6698.55
6698.55
LeanGate
15.58×
2.56
18.18
532.05
461.41
993.46
EuRoC MAV
DPV-SLAM [ 58 ]
2×
2.4
122.15
—
—
—
TABLE III: Core results across datasets comparing trajectory accuracy and computational efficiency. For Droid-SLAM, we report both single-GPU and parallel-mode profiling on the same device using official settings.
Category
Configuration
MAE ↓
RMSE ↓
Model Design
Default
baseline (dec6)
0.0281
0.0400
Head Design
w/o iterative head
0.0352
0.0488
Decoder Depth
dec6
0.0303
0.0430
Decoder Depth
dec3
0.0347
0.0492
Model Training
TABLE IV : Ablation on the ScanNet++ validation set for utility-score regression. The model design block evaluates architecture choices (iterative head and decoder depth), and the model training evaluates optimization choices (decoder freezing and pre-training). We report MAE and RMSE (lower is better); bold indicates the best value in each metric.
Fig. 6: Qualitative 3D trajectory comparisons on TUM-RGBD (fr1-teddy, fr1-room), 7-Scenes (heads-seq01), and EuRoC (MH01 easy, MH03 medium, MH04 difficult). Black denotes ground truth; orange shows Sim(3)-aligned estimates from LeanGate-filtered RGB streams; blue shows Sim(3)-aligned stride-15 results selected to match a similar retained-frame budget, illustrating our method’s tracking consistency under aggressive frame pruning.
Autonomous Intelligent Systems group, Computer Science Institute VI – Intelligent Systems and Robotics – and the Center for Robotics and the Lamarr Institute for Machine Learning and Artificial Intelligence, University of Bonn, Germany