Research map / Detection and segmentation / Tracking
PolarMOT: How Far Can Geometric Relations Take Us in 3D Multi-Object Tracking?
Aleksandr Kim, Guillem Brasó, Aljoša Ošep, Laura Leal-Taixé
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 time normalization to polar edge features (ablation) | PolarMOT · nuScenes val (mini training split) | AMOTA: 52.17 -> 57.96 | Improved |
| use localized polar instead of global Cartesian edge features (ablation) | PolarMOT · nuScenes val (mini training split) | AMOTA: 40.41 -> 57.96 | Improved |
| add time normalization to polar edge features (ablation) | PolarMOT · nuScenes val (full training set) | AMOTA: 70.27 -> 71.14 | Improved |
| use localized polar instead of Cartesian features (ablation) | PolarMOT · nuScenes val (full training set) | AMOTA: 69.75 -> 71.14 | Improved |
| add intra-frame spatial edges (ablation) | PolarMOT · nuScenes val | AMOTA: 70.09 -> 71.14 | Improved |
| add intra-frame spatial edges (ablation) | PolarMOT · nuScenes val | IDs: 198 -> 213 | Hurt |
| reduce max edge distance to 0.5x of measured max velocity (ablation) | PolarMOT online · nuScenes val | AMOTA: 67.27 -> 58.61 | Hurt |
| increase max edge distance to 2.0x (ablation) | PolarMOT online · nuScenes val | AMOTA: 67.27 -> 65.42 | Hurt |
| use prune+skip connectivity instead of consecutive chaining (ablation) | PolarMOT online · nuScenes val | AMOTA: 66.03 -> 67.27 | Improved |
| use prune+skip connectivity instead of dense linking (ablation) | PolarMOT online · nuScenes val | AMOTA: 61.39 -> 67.27 | Improved |
| aggregate past/present/future messages separately instead of spatial/temporal (ablation) | PolarMOT · nuScenes val | AMOTA: 55.83 -> 71.14 | Improved |
| aggregate past/present/future messages separately instead of all together (ablation) | PolarMOT · nuScenes val | AMOTA: 23.30 -> 71.14 | Improved |
Closest papers by meaning
- GNN3DMOT: Graph Neural Network for 3D Multi-Object Tracking With 2D-3D Multi-Feature Learning (CVPR 2020 · similarity 0.89)
- GRAE-3DMOT: Geometry Relation-Aware Encoder for Online 3D Multi-Object Tracking (CVPR 2025 · similarity 0.89)
- Seeing Behind Objects for 3D Multi-Object Tracking in RGB-D Sequences (CVPR 2021 · similarity 0.87)
- Object-Centric Representation Learning for Enhanced 3D Semantic Scene Graph Prediction (NeurIPS 2025 · similarity 0.87)
- Delving into Dynamic Scene Cue-Consistency for Robust 3D Multi-Object Tracking (AAAI 2026 · similarity 0.87)
- Towards Category Unification of 3D Single Object Tracking on Point Clouds (ICLR 2024 · similarity 0.86)
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.
