Detection & Tracking Demo
3D object detection (mAP/AP) and multi-object tracking (MOTA) evaluation on public RELLIS-3D-derived examples.
Detection Evaluation
| Candidate | mAP | Precision | Recall | F1 | Gate |
|---|
| Good |
1.0000 |
1.0000 |
1.0000 |
1.0000 |
PASS |
| Regressed |
0.1667 |
0.0000 |
0.0000 |
0.0000 |
FAIL |
Geometry: axis_aligned_3d_boxes · IoU thresholds: [0.25, 0.5] · Yaw ignored: True
Tracking Evaluation
| Candidate | MOTA | Recall | ID Switches | Mostly Tracked | Gate |
|---|
| Good |
1.0000 |
1.0000 |
0 |
1.0000 |
PASS |
| Regressed |
0.5556 |
0.7778 |
1 |
0.6667 |
PASS |
Geometry: axis_aligned_3d_boxes · IoU threshold: 0.5 · Yaw ignored: True
Detailed Reports
Build Locally
python scripts/build_detection_tracking_demo.py \
--output docs/demo/detection-tracking
Data derived from RELLIS-3D (CC BY-NC-SA 3.0).
Tracking examples are deterministic synthetic sequences.
Built with CloudAnalyzer.