Research map / Point clouds / Point cloud classification
Scene Flow Prior Based Point Cloud Completion with Masked Transformer (Student Abstract)
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes a self-supervised masked transformer for point cloud completion that uses forward-reverse scene flow (NSFP) from neighboring frames to generate mask-token proposals.
Problem
Point clouds from 3D sensors are sparse and incomplete, and single-frame completion methods learn from discontinuous datasets, limiting generalization to real-world tasks such as autonomous driving and tracking.
Approach
Use forward and reverse scene flow estimated between consecutive frames (Pt-1, Pt, Pt+1) as supplementary points for frame Pt, and feed them as mask tokens to a masked transformer (following Point-MAE-style proxies from Pang et al. 2022) so that multi-frame context constrains the completion.
Key results
- Reports Fidelity Distance of 0.01765 on the in-house pedestrian dataset.
- Reports Consistency score of 0.01097 (Chamfer distance between different outputs).
- Qualitative results show the method fills missing parts while preserving overall shape; no baseline comparisons reported.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use forward-reverse scene flow as a guide to generate token proposals | Scene Flow Prior Based Point Cloud Completion with Masked Transformer · Our pedestrian point cloud dataset | Consistency | Helped |
Limitations
- Evaluated only on a small in-house pedestrian dataset with no public benchmark.
- No baseline comparisons against existing point cloud completion methods.
- Dataset size and training details are sparsely reported (student abstract).
Experimental setup
Datasets: in-house pedestrian point cloud dataset
Benchmarks: in-house pedestrian point cloud dataset
Metrics: Fidelity Distance (Chamfer distance between input and output Consistency (Chamfer distance between different outputs)
Builds on: NSFP Point-MAE
Closest papers by meaning
- Point Cloud Semantic Scene Completion with Prototype-Guided Transformer (AAAI 2026 · similarity 0.88)
- Point Transformer V3: Simpler Faster Stronger (CVPR 2024 · similarity 0.86)
- PointCFormer: A Relation-Based Progressive Feature Extraction Network for Point Cloud Completion (AAAI 2025 · similarity 0.86)
- ComPC: Completing a 3D Point Cloud with 2D Diffusion Priors (ICLR 2025 · similarity 0.85)
- CRA-PCN: Point Cloud Completion with Intra- and Inter-level Cross-Resolution Transformers (AAAI 2024 · similarity 0.85)
- A Consistency-Aware Spot-Guided Transformer for Versatile and Hierarchical Point Cloud Registration (NeurIPS 2024 · similarity 0.84)
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.