Research map / Detection and segmentation / Tracking
GNN3DMOT: Graph Neural Network for 3D Multi-Object Tracking With 2D-3D Multi-Feature Learning
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Combine 2D and 3D appearance and motion features rather than using a single feature type or modality (ablation) | GNN3DMOT · KITTI-Car val | sAMOTA (%): 2D+3D M+A: 93.68; 2D A: 88.31; 2D M: 64.24; 3D A: 88.27; 3D M: 88.57; 2D+3D A: 89.39; 2D+3D M: 91.75; 2D M+A: 90.56; 3D | Helped |
| Concatenate the 2D and 3D features instead of adding them (ablation) | GNN3DMOT · KITTI-Car val | sAMOTA (%): 89.98 -> 93.68 | Helped |
| Use a two-layer MLP rather than negative L2 distance or cosine similarity for edge regression (ablation) | GNN3DMOT · KITTI-Car val | sAMOTA (%): 82.26; 87.07 -> 93.68 | Helped |
| Use affinity-weighted feature-difference aggregation (Type 4) rather than the other tested node aggregation rules (ablation) | GNN3DMOT · KITTI-Car val | sAMOTA (%): 75.61; 87.81; 89.78; 91.15; 91.66; 91.88; 92.17 -> 93.68 | Helped |
| Use ensemble branch dropping during training; the best tested drop ratio is r=0.5 (ablation) | GNN3DMOT · KITTI-Car val | 3D MOT accuracy | Helped |
| Increase GNN depth, with three GNN layers selected as best among the tested depths (ablation) | GNN3DMOT · KITTI-Car val | 3D MOT accuracy | Helped |
| Use later GNN-layer features rather than deactivating the GNN (ablation) | GNN3DMOT · KITTI-Car val | 3D MOT accuracy | Helped |
| Replace projected 3D detections with detections from a 2D detector | GNN3DMOT · KITTI-Car test | MOTA (%): 80.40 -> 82.24 | Helped |
Closest papers by meaning
- Multi-Scene Generalized Trajectory Global Graph Solver with Composite Nodes for Multiple Object Tracking (AAAI 2024 · similarity 0.88)
- GRAE-3DMOT: Geometry Relation-Aware Encoder for Online 3D Multi-Object Tracking (CVPR 2025 · similarity 0.88)
- S2-Track: A Simple yet Strong Approach for End-to-End 3D Multi-Object Tracking (ICML 2025 · similarity 0.87)
- Multiple Object Tracking as ID Prediction (CVPR 2025 · similarity 0.87)
- ADA-Track: End-to-End Multi-Camera 3D Multi-Object Tracking with Alternating Detection and Association (CVPR 2024 · similarity 0.87)
- M3SOT: Multi-Frame, Multi-Field, Multi-Space 3D Single Object Tracking (AAAI 2024 · similarity 0.87)
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.