Research map / Learning theory and algorithms / Matrix and tensor methods
Square Root Marginalization for Sliding-Window Bundle Adjustment
Nikolaus Demmel, David Schubert, Christiane Sommer, Daniel Cremers, Vladyslav Usenko
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add square root (NS+QR) marginalization instead of SC+SC (ablation) | square root VIO with SC+LDLT optimization, 32-bit · EuRoC VIO | ATE (m): 0.211 -> 0.068 | Improved |
| add square root (NS+QR) marginalization instead of SC+SC (ablation) | square root VIO with NS+LDLT optimization, 32-bit · EuRoC VIO | ATE (m): 0.232 -> 0.068 | Improved |
| factorize SC prior into square root form (LDLT) only (ablation) | SC marginalization, 32-bit · EuRoC VIO | ATE (m): 0.211 | Hurt |
| add nullspace-projection (NS+LDLT) optimization instead of SC+LDLT (ablation) | square root VIO with NS+QR marginalization, 32-bit · EuRoC VIO | total runtime (s): 21.8 -> 14.9 | Improved |
| add nullspace-projection (NS+LDLT) optimization instead of SC+LDLT (ablation) | square root VIO with NS+QR marginalization, 32-bit · EuRoC VIO | ATE (m): 0.068 -> 0.068 | No clear effect |
| switch from 64-bit to 32-bit floats | square root VIO · EuRoC VIO | total runtime (s): 17.9 -> 14.9 | Improved |
| switch from 64-bit to 32-bit floats | squared (SC) VIO · eurocMH01 | ATE (m): 0.093 -> 0.991 | Hurt |
| switch from 64-bit to 32-bit floats | squared (SC) VO · kitti10 | ATE (m): 1.11 -> 26.12 | Hurt |
| switch from 64-bit to 32-bit floats | NS-projection vs SC optimization · VIO optimization runtime | speedup: 22% vs 8% | Improved |
| use square root VIO 32-bit instead of squared VIO 64-bit baseline | Basalt-based estimator · VIO total runtime | speedup: 36% | Improved |
| use square root QR marginalization instead of SC | VIO 32-bit · eurocMH01 | marginalization runtime (s): 1.7 -> 2.3 | Hurt |
| use square root marginalization and NS optimization instead of SC+SC and SC+LDLT (ablation) | square root VO, 32-bit · KITTI VO | ATE (m): 3.479 -> 3.216 | Improved |
Closest papers by meaning
- From Variance to Veracity: Unbundling and Mitigating Gradient Variance in Differentiable Bundle Adjustment Layers (CVPR 2024 · similarity 0.86)
- Square Root Bundle Adjustment for Large-Scale Reconstruction (CVPR 2021 · similarity 0.85)
- EqNIO: Subequivariant Neural Inertial Odometry (ICLR 2025 · similarity 0.84)
- Direct Sparse Odometry (TPAMI 2017 · similarity 0.83)
- Adaptive VIO: Deep Visual-Inertial Odometry with Online Continual Learning (CVPR 2024 · similarity 0.82)
- Leveraging Consistent Spatio-Temporal Correspondence for Robust Visual Odometry (AAAI 2025 · similarity 0.81)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
