Research map / Detection and segmentation / 3D detection and driving
Once Detected, Never Lost: Surpassing Human Performance in Offline LiDAR based 3D Object Detection
Lue Fan, Yuxue Yang, Yiming Mao, Feng Wang, Yuntao Chen, Naiyan Wang, Zhaoxiang Zhang
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 bidirectional track extension (backtracing) on top of forward-only extension (ablation) | CTRL refinement · WOD val Vehicle | T-FN ratio: 9.1k (0.87%) -> 5.1k (0.48%) | Improved |
| add track-centric refinement module to offline FSD (ablation) | Offline FSD · WOD val Vehicle L2 | APH: 74.4 -> 80.1 | Improved |
| add bidirectional tracking extension before refinement (ablation) | CTRL · WOD val Vehicle L2 | APH: 80.1 -> 81.0 | Improved |
| bidirectional tracking adds boxes without adding high-confidence false positives (ablation) | CTRL · WOD val Vehicle | H-FP: 8.1k (0.77%) -> 8.1k (0.77%) | No clear effect |
| do not cap input track length (use full length) (ablation) | track-centric learning · WOD val Vehicle L2 | APH: 77.1 -> 80.1 | Improved |
| use track-centric instead of object-centric label assignment (ablation) | track-centric learning · WOD val Vehicle L2 | APH: 79.6 -> 80.1 | Improved |
| separate static and dynamic tracks into two pipelines (ablation) | track-centric learning · WOD val Vehicle L2 | APH: 80.1 -> 79.8 | No clear effect |
| use MIMO instead of MISO (ablation) | track-centric learning · WOD val Vehicle L2 | APH / FPS: 79.5 -> 80.1; 1.1 -> 33.6 FPS | Improved |
| give MISO a 4x longer training schedule (ablation) | MISO · WOD val Vehicle L2 | APH: 79.5 -> 80.0 | Improved |
| add track coherence optimization given one GT box (ablation) | CTRL · WOD val Vehicle L1 3D AP IoU=0.8 | 3D AP: 64.9 -> 76.5 | Improved |
| add track TTA (ablation) | CTRL · WOD val | mAP/mAPH L2: 83.6/82.1 -> 83.9/82.3 | Improved |
| replace 3DAL with CTRL for offline detection | offline auto-labeling · WOD val Vehicle 3D AP strict | 3D AP: 57.8 -> 64.9 | Improved |
Closest papers by meaning
- Detecting As Labeling: Rethinking LiDAR-camera Fusion in 3D Object Detection (ECCV 2024 · similarity 0.91)
- TCC-Det: Temporarily consistent cues for weakly-supervised 3D detection (ECCV 2024 · similarity 0.91)
- Towards A Weakly Supervised Framework for 3D Point Cloud Object Detection and Annotation (TPAMI 2021 · similarity 0.90)
- iDet3D: Towards Efficient Interactive Object Detection for LiDAR Point Clouds (AAAI 2024 · similarity 0.89)
- MOS: Model Synergy for Test-Time Adaptation on LiDAR-Based 3D Object Detection (ICLR 2025 · similarity 0.89)
- A-Teacher: Asymmetric Network for 3D Semi-Supervised Object Detection (CVPR 2024 · similarity 0.89)
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.
