Research map / Point clouds / Shape and surface completion
3D MeanFlow: One-Step Point Cloud Completion and Generation via Average-Velocity Transport
Haowen Zhong, Jiujun Cheng, Haowen Wang, Chao Wei, Lu Yang, Qichao Mao, Shangce Gao
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 shape-level consistency term (lambda 2.0 vs 0) (ablation) | 3DMF · ShapeNet Car completion, 1NFE | CD: 1.529 -> 1.224 | Improved |
| add shape-level consistency term (lambda 2.0 vs 0) (ablation) | 3DMF · ShapeNet Car completion, 5NFE | CD: 1.293 -> 0.975 | Improved |
| raise shape consistency weight from 2.0 to 5.0 (ablation) | 3DMF · ShapeNet Car completion, 1NFE | CD: 1.224 -> 1.431 | Hurt |
| replace uniform(0,1) time sampler by beta(2,2) with r~uniform(0.65t,t) (ablation) | 3DMF · ShapeNet Car completion, 1NFE | CD: 3.525 -> 1.224 | Improved |
| replace logit-normal time sampler by beta(2,2) with r~uniform(0.65t,t) (ablation) | 3DMF · ShapeNet Car completion, 1NFE | CD: 2.471 -> 1.224 | Improved |
| use beta(2,5) instead of beta(2,2) high-t mixture (ablation) | 3DMF · ShapeNet Car completion, 1NFE | CD: 1.224 -> 1.473 | Hurt |
| narrow r interval to uniform(0.65t,t) vs uniform(0.5t,t) (ablation) | 3DMF · ShapeNet Car completion, 1NFE | EMD: 1.896 vs 1.861 | No clear effect |
| use 5 steps instead of 1 step | 3DMF · ShapeNet Car completion | CD: 1.224 -> 0.975 | Improved |
| use 5 steps instead of 1 step | 3DMF · ShapeNet Airplane completion | CD: 0.639 -> 0.369 | Improved |
| use 5 steps instead of 1 step | 3DMF · ShapeNet Airplane generation | 1-NNA CD: 76.19 -> 73.38 | Improved |
| insert PointPlug one-step completion into FSHNet | FSHNet · nuScenes val | mAP: 68.1 -> 68.5 | Improved |
| insert PointPlug into FSHNet | FSHNet · nuScenes val Car mAP 50m+ | Car mAP: 16.23 -> 23.85 | Improved |
Closest papers by meaning
- Fast Point Cloud Generation With Straight Flows (CVPR 2023 · similarity 0.90)
- Points-to-3D: Structure-Aware 3D Generation with Point Cloud Priors (CVPR 2026 · similarity 0.89)
- PointMAC: Meta-Learned Adaptation for Robust Test-Time Point Cloud Completion (NeurIPS 2025 · similarity 0.88)
- PointInfinity: Resolution-Invariant Point Diffusion Models (CVPR 2024 · similarity 0.87)
- GenPC: Zero-shot Point Cloud Completion via 3D Generative Priors (CVPR 2025 · similarity 0.87)
- SuperPC: A Single Diffusion Model for Point Cloud Completion, Upsampling, Denoising, and Colorization (CVPR 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.
