Research map / 3D vision / 3D generation
Distilling Diffusion Models to Efficient 3D LiDAR Scene Completion
Shengyuan Zhang, An Zhao, Ling Yang, Zejian Li, Chenye Meng, Haoran Xu, Tianrun Chen, AnYang Wei, Perry Pengyun Gu, Lingyun Sun
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 structural loss in distillation (ablation) | ScoreLiDAR (Refined) · SemanticKITTI | CD: 0.419 -> 0.342 | Improved |
| add structural loss in distillation (ablation) | ScoreLiDAR (Refined) · KITTI-360 | CD: 0.549 -> 0.452 | Improved |
| add scene-wise loss to structural loss (ablation) | ScoreLiDAR · SemanticKITTI | CD: 0.367 -> 0.342 | Improved |
| add point-wise loss to structural loss (ablation) | ScoreLiDAR · SemanticKITTI | CD: 0.351 -> 0.342 | Improved |
| lower scene weight to 0.05 from 0.5 (ablation) | ScoreLiDAR · KITTI-360 | CD: 0.452 -> 0.494 | Hurt |
| add structural loss to LiDiff training (ablation) | LiDiff (Refined) · SemanticKITTI | CD: 0.375 -> 0.399 | Hurt |
| distill LiDiff into ScoreLiDAR (5.37 s vs 30.55 s) | LiDiff (Refined) · SemanticKITTI | CD: 0.375 -> 0.342 | Improved |
| use ScoreLiDAR distillation instead of fewer LiDiff steps (8 steps) | LiDiff 8 steps Refined · SemanticKITTI | CD: 0.411 -> 0.342 | Improved |
| reduce sampling steps from 8 to 4 (ablation) | ScoreLiDAR (Refined) · SemanticKITTI | CD: 0.342 -> 0.326 | Improved |
| reduce sampling steps from 8 to 1 (ablation) | ScoreLiDAR (Refined) · SemanticKITTI | CD: 0.342 -> 0.548 | Hurt |
| use curvature-based key point selection instead of random (ablation) | ScoreLiDAR · SemanticKITTI | CD: 0.384 -> 0.342 | Improved |
| use curvature-based key point selection instead of farthest (ablation) | ScoreLiDAR · SemanticKITTI | CD: 0.442 -> 0.342 | Improved |
Closest papers by meaning
- Scaling Diffusion Models to Real-World 3D LiDAR Scene Completion (CVPR 2024 · similarity 0.93)
- Towards Realistic Scene Generation with LiDAR Diffusion Models (CVPR 2024 · similarity 0.89)
- Distilling Monocular Foundation Model for Fine-grained Depth Completion (CVPR 2025 · similarity 0.87)
- Three Pillars Improving Vision Foundation Model Distillation for Lidar (CVPR 2024 · similarity 0.87)
- Skip Mamba Diffusion for Monocular 3D Semantic Scene Completion (AAAI 2025 · similarity 0.86)
- Towards Learning to Complete Anything in Lidar (ICML 2025 · similarity 0.86)
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.
