Research map / Detection and segmentation / 3D detection and driving
Joint Monocular 3D Vehicle Detection and Tracking
Hou-Ning Hu, Qi-Zhi Cai, Dequan Wang, Ji Lin, Min Sun, Philipp Krahenbuhl, Trevor Darrell, Fisher Yu
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 deep feature correlation to association (ablation) | KF3D tracker · GTA tracking | MOTA / FN: 69.616 -> 69.843 MOTA; FN 21.298 -> 18.547 | Improved |
| add depth-order matching and occlusion-aware association (ablation) | KF3D tracker without deep feature · GTA tracking | mismatches (MM): 17480 -> 16042 | Improved |
| add depth-order matching and occlusion-aware association (ablation) | KF3D tracker with deep feature · GTA tracking | mismatches (MM): 27639 -> 25860 | Improved |
| replace Kalman filter with LSTM motion model (ablation) | full tracker · GTA tracking | MOTA: 70.126 -> 70.439 | Improved |
| replace Kalman filter with LSTM motion model (ablation) | 3D estimation · GTA 3D IoU AP (Easy, AP70) | 3D IoU AP70: 6.14 -> 7.89 | Improved |
| add Kalman filter smoothing over single-frame estimation (ablation) | 3D estimation · GTA 3D IoU AP (Easy, AP70) | 3D IoU AP70: 6.13 -> 6.14 | No clear effect |
| use 3D motion model instead of 2D (ablation) | tracker without deep feature or order · GTA tracking (100m) | MOTA: 57.042 -> 69.616 | Improved |
| predict projection of 3D box center instead of using 2D box center (ablation) | tracking pipeline · KITTI training set | IDS / MOTA: IDS 315 -> 190; MOTA 91.06 -> 91.58 | Improved |
| increase depth-module training data from 1% to 100% (ablation) | 3D box estimation · GTA | Abs Rel: 0.258 -> 0.112 | Improved |
| use ground-truth camera pose instead of Mono-VO pose (ablation) | tracker · KITTI | MOTA: 71.079 -> 71.215 | No clear effect |
| add location-based matching and reduce max-age from 30 to 20 (ablation) | tracker · KITTI (appendix) | MOTA: 89.93 -> 90.97 | Improved |
| monocular 3D tracker versus LiDAR-centric baseline at 30m | proposed tracker · Argoverse val (30m) | MOTA: 73.02 -> 77.93 | Improved |
Closest papers by meaning
- Monocular Quasi-Dense 3D Object Tracking (TPAMI 2022 · similarity 0.92)
- Time3D: End-to-End Joint Monocular 3D Object Detection and Tracking for Autonomous Driving (CVPR 2022 · similarity 0.91)
- Argoverse: 3D Tracking and Forecasting With Rich Maps (CVPR 2019 · similarity 0.89)
- A Depth Complementary Framework for Monocular 3D Object Detection in Autonomous Driving (TPAMI 2026 · similarity 0.89)
- Kinematic 3D Object Detection in Monocular Video (ECCV 2020 · similarity 0.88)
- TrajectoryFormer: 3D Object Tracking Transformer with Predictive Trajectory Hypotheses (ICCV 2023 · similarity 0.88)
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.
