Research map / Point clouds / Point cloud classification
Probabilistic Discrepancy Learning for Roadside LiDAR Scene Completion
Xiaogang Wu, Jinchao Hu, Zixian Wang, Dun Liu, BoXiang Cheng, Yiqiang Wu
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 the global sampling discrepancy loss to the diffusion-only baseline (ablation) | PDL variant A · V2X-Seq | Chamfer distance: 0.39 -> 0.37 | Improved |
| add the region discrepancy learning loss on top of the global loss (ablation) | PDL variant B · V2X-Seq | Chamfer distance: 0.37 -> 0.34 | Improved |
| add PPDM pose correction on top of both losses (ablation) | PDL variant C · V2X-Seq | Chamfer distance: 0.34 -> 0.29 | Improved |
| add PPDM pose correction on top of both losses (ablation) | PDL variant C · V2X-Seq | 3D JSD: 0.53 -> 0.48 | Improved |
| add PPDM pose correction on top of both losses (ablation) | PDL variant C · V2X-Seq (voxel 0.5) | Voxel IoU: 54.7 -> 55.9 | Improved |
| add PPDM correction to BEVHeight detections (Easy) | BEVHeight · V2X-Seq val, Car 3D AP@0.5 Easy | 3D AP: 77.78 -> 90.32 | Improved |
| add PPDM correction to BEVHeight detections (Moderate) | BEVHeight · V2X-Seq val, Car 3D AP@0.5 Moderate | 3D AP: 65.77 -> 86.78 | Improved |
| add PPDM correction to BEVHeight detections (Hard) | BEVHeight · V2X-Seq val, Car 3D AP@0.5 Hard | 3D AP: 65.85 -> 84.26 | Improved |
| use PDL (pose correction plus diffusion with discrepancy losses) instead of LiDPM | diffusion scene completion · V2X-Seq | Chamfer distance: 0.35 -> 0.29 | Improved |
| use PDL instead of LiDPM on an unseen dataset without fine-tuning | diffusion scene completion · TUMTraf-V2X (cross-dataset) | Chamfer distance: 0.34 -> 0.30 | Improved |
| use PDL instead of LiDPM on an unseen dataset | diffusion scene completion · TUMTraf-V2X (voxel 0.5) | Voxel IoU: 54.9 -> 55.4 | Improved |
Closest papers by meaning
- Learnability-Driven Submodular Optimization for Active Roadside 3D Detection (CVPR 2026 · similarity 0.86)
- LiDAR-PTQ: Post-Training Quantization for Point Cloud 3D Object Detection (ICLR 2024 · similarity 0.86)
- Exploring Scene Affinity for Semi-Supervised LiDAR Semantic Segmentation (CVPR 2025 · similarity 0.85)
- Effective Multi-sensor Conditioning for Street-view Novel-view Synthesis (NeurIPS 2026 · similarity 0.85)
- Scaling Diffusion Models to Real-World 3D LiDAR Scene Completion (CVPR 2024 · similarity 0.85)
- LidaRF: Delving into Lidar for Neural Radiance Field on Street Scenes (CVPR 2024 · similarity 0.85)
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.
