Research map / Point clouds / Shape and surface completion
Learning Manifolds in High-D Point Embedding for Anisotropic Surface Approximation from Unstructured Point Clouds
Hongbo Li, Haikuan Zhu, Xiaohu Guo, Wenping Wang, Jing Hua, Zichun Zhong
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 |
|---|---|---|---|
| use neighborhood distance-based high-d embedding loss with MAE instead of plain MAE (ablation) | HD-PEA · Thingi10K (loss ablation) | Chamfer distance (x1e-5): 29.737 -> 0.496 | Improved |
| use neighborhood distance-based loss instead of plain MSE (ablation) | HD-PEA · Thingi10K (loss ablation) | Chamfer distance (x1e-5): 30.864 -> 30.072 | Improved |
| use neighborhood distance-based loss with MSE instead of plain MSE (ablation) | HD-PEA · Thingi10K (loss ablation) | F1: 0.953 -> 0.984 | Improved |
| use MAE with neighborhood loss instead of MSE (ablation) | HD-PEA · Thingi10K (loss ablation) | Hausdorff distance (x1e-2): 1.065 -> 0.678 | Improved |
| add patch-based meta-embedding inference (PMEI) (ablation) | HD-PEA · Myles et al. complicated model | visual quality | Improved |
| estimate tangent space via PCA on 3D neighbors instead of high-d neighbors (ablation) | HD-PEA · tangent space approximation | topology preservation | Improved |
| add 0.5% noise to real scan (ablation) | HD-PEA · Myles et al. synthetic scans (75 models) | Chamfer distance (x1e-5): 0.649 -> 1.417 | Hurt |
| add 0.1% normal noise (ablation) | HD-PEA · Myles et al. synthetic scans (75 models) | Chamfer distance (x1e-5): 0.649 -> 0.673 | No clear effect |
| add 0.2% normal noise (ablation) | HD-PEA · Myles et al. synthetic scans (75 models) | Chamfer distance (x1e-5): 0.649 -> 0.674 | No clear effect |
| use irregular striped sampling (ablation) | HD-PEA · Myles et al. synthetic scans (75 models) | Chamfer distance (x1e-5): 0.649 -> 0.688 | No clear effect |
| increase input points from 20,000 to 60,000 (ablation) | HD-PEA · Myles et al. | Chamfer distance (x1e-5): 0.648 -> 0.628 | No clear effect |
| add NASM remeshing after NDC | NDC (128^3) · Myles et al. synthetic scans (75 models) | Chamfer distance (x1e-5): 0.749 -> 0.792 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- POCO (64^3) — POCO: Point Convolution for Surface Reconstruction (CVPR 2022) · better on 6 of 6 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- GeoUDF: Surface Reconstruction from 3D Point Clouds via Geometry-guided Distance Representation (ICCV 2023 · similarity 0.89)
- Meshing Point Clouds with Predicted Intrinsic-Extrinsic Ratio Guidance (ECCV 2020 · similarity 0.88)
- High-Fidelity Lightweight Mesh Reconstruction from Point Clouds (CVPR 2025 · similarity 0.88)
- Arbitrary-Scale Point Cloud Upsampling by Voxel-Based Network with Latent Geometric-Consistent Learning (AAAI 2024 · similarity 0.87)
- Dual Manifold Regularization Steered Robust Representation Learning for Point Cloud Analysis (AAAI 2025 · similarity 0.87)
- A Lightweight UDF Learning Framework for 3D Reconstruction Based on Local Shape Functions (CVPR 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.
