Research map / Point clouds / Shape and surface completion
Revisiting Point Cloud Completion: Are We Ready For The Real-World?
Stuti Pathak, Prashant Kumar, Dheeraj Baiju, Nicholus Mboga, Gunther Steenackers, Rudi Penne
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 0-dim persistent homology topological loss on decoder seeds (ablation) | ODGNet · RealPC completion | CD-L1 (x1e-3): 119 -> 103 | Improved |
| add 0-dim persistent homology topological loss on decoder seeds (ablation) | ODGNet · RealPC completion | CD-L2 (x1e-3): 111 -> 80 | Improved |
| replace PH computation with Homology Sampler (BOSH) proxy backbones | ODGNet / TopoODG · RealPC completion | CD-L2 (x1e-3): 111 -> 5.4 | Improved |
| replace PH computation with Homology Sampler (BOSH) proxy backbones | SnowflakeNet · RealPC completion | CD-L1 (x1e-3): 60 -> 69 | No clear effect |
| add 0-dim PH priors to DGCNN-based generative model (ablation) | DGCNN-based model · KITTI scene completion | Chamfer distance (x1e-3): 1.18 -> 1.04 | Improved |
| add 0-dim PH priors to DGCNN-based generative model (ablation) | DGCNN-based model · KITTI scene reconstruction | Chamfer distance (x1e-3): 1.50 -> 1.18 | Improved |
| evaluate completion on real-world industrial point clouds instead of PCN | ODGNet · completion mean CD-L1 (x1e-3) | CD-L1 (x1e-3): 6 -> 119 | Hurt |
| evaluate completion on real-world industrial point clouds instead of PCN | AdaPoinTr · completion mean CD-L1 (x1e-3) | CD-L1 (x1e-3): 7 -> 69 | Hurt |
| evaluate simplification on RealPC instead of ShapeNet | WLOP simplification · non-neural simplification | Chamfer distance (x1e-3): 0.522 -> 8.817 | Hurt |
| evaluate alpha-shape surface reconstruction on RealPC instead of ShapeNet | alpha shapes · non-neural surface reconstruction | Hausdorff distance (x1e-3): 68.198 -> 494.591 | Hurt |
| evaluate reconstruction on RealPC instead of ShapeNet | FoldingNet · reconstruction | Chamfer distance (x1e-3): 5 -> 27 | Hurt |
| evaluate diffusion shape generation on RealPC instead of ShapeNet | LION · shape generation 1-NNA (CD) | 1-NNA: 67 -> 90 | Hurt |
Closest papers by meaning
- LAKe-Net: Topology-Aware Point Cloud Completion by Localizing Aligned Keypoints (CVPR 2022 · similarity 0.91)
- On the Effectiveness of Persistent Homology (NeurIPS 2022 · similarity 0.90)
- Point-Level Topological Representation Learning on Point Clouds (ICML 2025 · similarity 0.89)
- The Flood Complex: Large-Scale Persistent Homology on Millions of Points (NeurIPS 2025 · similarity 0.89)
- Rethinking Point Cloud Data Augmentation: Topologically Consistent Deformation (ICML 2025 · similarity 0.87)
- SymmCompletion: High-Fidelity and High-Consistency Point Cloud Completion with Symmetry Guidance (AAAI 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.
