Research map / Detection and segmentation / 3D detection and driving
When the City Teaches the Car: Label-Free 3D Perception from Infrastructure
ZHEN XU, Jinsu Yoo, Cristian Bautista, Zanming Huang, Tai-Yu Pan, Zhenzhen Liu, Katie Luo, Mark Campbell, Bharath Hariharan, Wei-Lun Chao
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 tracking-based box refinement to PP-score pseudo-labels (ablation) | RSU pseudo-labels, CenterPoint ego · Town10 ego test | average AP: 72.7 -> 76.7 | Improved |
| add tracking-based refinement (ablation) | RSU pseudo-labels, PointPillars ego · Town10 ego test | average AP: 73.1 -> 78.6 | Improved |
| simulate communication noise in RSU-to-ego broadcasting (ablation) | PP score + tracking, CenterPoint · Town10 ego test | average AP: 76.7 -> 66.4 | Hurt |
| add heuristic box refinement under communication noise (ablation) | PP score + tracking, CenterPoint · Town10 ego test | average AP: 66.4 -> 69.8 | Improved |
| add heuristic box refinement under communication noise (ablation) | PP score + tracking, CenterPoint · Town10 ego test (Car AP) | Car AP: 80.4 -> 77.7 | Hurt |
| remove box filtering of distant or sparse boxes (ablation) | pseudo-label aggregation · Town10 ego test | average AP: 60.0 -> 66.4 | Improved |
| use supervised RSU detectors instead of unsupervised | CenterPoint ego · Town10 ego test | average AP: 76.7 -> 90.6 | Improved |
| scale to 4 towns and 48 RSUs | CenterPoint ego · combined 4-town test | average AP: 80.7 vs 76.7 (single town) | No clear effect |
| fine-tune rural-trained ego detector with urban infra pseudo-labels | CenterPoint ego · Town10 urban test | average AP: 64.3 -> 79.3 | Improved |
| combine infra pseudo-labels with Oyster pseudo-labels | PointRCNN · dynamic-object AP | Dyn AP: 62.0 -> 64.9 | Improved |
| combine infra pseudo-labels with Oyster pseudo-labels under comm. noise | PointRCNN · dynamic-object AP | Dyn AP: 49.9 -> 59.5 | Improved |
| increase RSUs from 4 or 8 to 12 | ego detector · Town10 ego test | AP | Improved |
Closest papers by meaning
- Learning 3D Perception from Others' Predictions (ICLR 2025 · similarity 0.91)
- Unsupervised 3D Perception with 2D Vision-Language Distillation for Autonomous Driving (ICCV 2023 · similarity 0.88)
- Learning to Drive is a Free Gift: Large-Scale Label-Free Autonomy Pretraining from Unposed In-The-Wild Videos (CVPR 2026 · similarity 0.88)
- Transfer Your Perspective: Controllable 3D Generation from Any Viewpoint in a Driving Scene (CVPR 2025 · similarity 0.87)
- Prioritizing Perception-Guided Self-Supervision: A New Paradigm for Causal Modeling in End-to-End Autonomous Driving (NeurIPS 2025 · similarity 0.87)
- UrbanIng-V2X: A Large-Scale Multi-Vehicle, Multi-Infrastructure Dataset Across Multiple Intersections for Cooperative Perception (NeurIPS 2025 · 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.
