Research map / 3D vision / Depth and stereo
ClusterSLAM: A SLAM Backend for Simultaneous Rigid Body Clustering and Motion Estimation
Jiahui Huang, Sheng Yang, Zishuo Zhao, Yu-Kun Lai, Shi-Min Hu
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 noise-aware (Mahalanobis, weighted) motion distance (ablation) | ClusterSLAM · Synthetic indoor sequences | RMSE (m): 1.21 -> 0.44 | Improved |
| add noise-aware (Mahalanobis, weighted) motion distance (ablation) | ClusterSLAM · Synthetic outdoor sequences | clustering accuracy (%): 86.51 -> 94.15 | Improved |
| add image-space prior term to motion distance (ablation) | ClusterSLAM · Synthetic indoor sequences | clustering accuracy (%): 76.16 -> 91.54 | Improved |
| add image-space prior term to motion distance (ablation) | ClusterSLAM · Synthetic outdoor sequences | RMSE (m): 9.56 -> 0.63 | Improved |
| use noise-aware ICP instead of point-to-point ICP for cluster initialization (ablation) | ClusterSLAM · Synthetic outdoor sequences | RMSE (m): 8.54 -> 0.94 | Improved |
| use noise-aware ICP instead of point-to-point ICP for cluster initialization (ablation) | ClusterSLAM · Synthetic indoor sequences | RMSE (m): 0.22 -> 0.12 | Improved |
| use noise-aware ICP instead of point-to-point ICP (ablation) | ClusterSLAM · Synthetic outdoor sequences | initialization time (ms): 0.01 -> 0.12 | Hurt |
| use decoupled instead of fully-coupled optimization (ablation) | ClusterSLAM · Synthetic outdoor sequences | RMSE (m): 0.73 -> 0.12 | Improved |
| use decoupled instead of fully-coupled optimization (ablation) | ClusterSLAM · Synthetic indoor sequences | RMSE (m): 0.034 -> 0.047 | Hurt |
| use decoupled instead of semi-decoupled optimization (ablation) | ClusterSLAM · Synthetic indoor sequences | time per iteration (s): 1.04 -> 0.57 | Improved |
| use hierarchical agglomerative clustering instead of affinity propagation or spectral clustering (ablation) | ClusterSLAM · Motion distance matrix 260x260 | running time: 13ms -> 6.3ms (AP), 180ms -> 6.3ms (SC) | Improved |
| use smaller chunk sizes (30 to 300 range) (ablation) | ClusterSLAM · CARLA long sequences | beta VI and clustering time | Mixed |
Closest papers by meaning
- ClusterVO: Clustering Moving Instances and Estimating Visual Odometry for Self and Surroundings (CVPR 2020 · similarity 0.88)
- DyGS-SLAM: Real-Time Accurate Localization and Gaussian Reconstruction for Dynamic Scenes (ICCV 2025 · similarity 0.87)
- Finding Geometric Models by Clustering in the Consensus Space (CVPR 2023 · similarity 0.87)
- Robust Multi-Body Feature Tracker: A Segmentation-Free Approach (CVPR 2016 · similarity 0.86)
- DG-SLAM: Robust Dynamic Gaussian Splatting SLAM with Hybrid Pose Optimization (NeurIPS 2024 · similarity 0.86)
- ProDyG: Progressive Dynamic Scene Reconstruction via Gaussian Splatting from Monocular Videos (NeurIPS 2025 · similarity 0.85)
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.
