Research map / Point clouds / Shape and surface completion
Unsupervised Inference of Signed Distance Functions From Single Sparse Point Clouds Without Learning Priors
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Learn surface parameterization and SDF inference end-to-end (vs. optimizing separately, using only the 5000 predicted points) (ablation) | NeuralTPS · ShapeNet airplane 300 points | CD_L1 x10: 0.105 -> 0.095 | Helped |
| Stop gradient from SDF inference loss into surface parameterization (vs. allowing backprop) (ablation) | NeuralTPS · ShapeNet airplane 300 points | CD_L2 x100: 0.047 -> 0.030 | Helped |
| Replace learned surface parameterization with a point cloud upsampling method (ablation) | NeuralTPS · ShapeNet airplane 300 points | CD_L1 x10: 0.095 -> 0.582 | Hurt |
| Use single chart instead of multiple patches (AtlasNet-style 3 or 5 patches) (ablation) | NeuralTPS · ShapeNet airplane 300 points | CD_L1 x10: 0.095 -> 0.114 (3 patches), 0.115 (5 patches) | Helped |
| Remove Chamfer loss L_CD on surface parameterization (ablation) | NeuralTPS · ShapeNet airplane 300 points | CD_L1 x10: 0.146 -> 0.095 | Helped |
| Add surface constraint loss L_Surf (ablation) | NeuralTPS · ShapeNet airplane 300 points | CD_L1 x10: 0.108 -> 0.095 | Helped |
| Use thin plate spline basis r^2 log r instead of |r|^3 (ablation) | NeuralTPS · ShapeNet airplane 300 points | CD_L1 x10: 0.110 -> 0.095 | Helped |
| Perform TPS in learned feature space instead of spatial space (ablation) | NeuralTPS · ShapeNet airplane 300 points | CD_L1 x10: 0.791 -> 0.095 | Helped |
| Add displacement MLP3 predicting d_FEA (ablation) | NeuralTPS · ShapeNet airplane 300 points | CD_L1 x10: 0.159 -> 0.095 | Helped |
| Add 1-3% Gaussian input noise (ablation) | NeuralTPS · ShapeNet airplane 300 points | CD_L1 x10: 0.095 -> 0.103 (1%), 0.177 (2%), 0.272 (3%) | Hurt |
| Proposed method vs. state-of-the-art on ShapeNet 300-point shapes | NeuralTPS · ShapeNet 300 points | CD_L1 x10 / CD_L2 x100 / NC: CD_L1 mean 0.141 vs next best 0.183 (NPull); NC 0.899 vs 0.883 | Helped |
| Proposed method vs. state-of-the-art on 3D Scene | NeuralTPS · 3D Scene | CD_L1 mean: 0.077 vs 0.097 (NPull) | Helped |
Closest papers by meaning
- Learning Bijective Surface Parameterization for Inferring Signed Distance Functions from Sparse Point Clouds with Grid Deformation (CVPR 2025 · similarity 0.93)
- Inferring Neural Signed Distance Functions by Overfitting on Single Noisy Point Clouds through Finetuning Data-Driven based Priors (NeurIPS 2024 · similarity 0.93)
- Toward Robust Neural Reconstruction from Sparse Point Sets (CVPR 2025 · similarity 0.90)
- MultiPull: Detailing Signed Distance Functions by Pulling Multi-Level Queries at Multi-Step (NeurIPS 2024 · similarity 0.89)
- Unsupervised Occupancy Learning from Sparse Point Cloud (CVPR 2024 · similarity 0.89)
- A Lightweight UDF Learning Framework for 3D Reconstruction Based on Local Shape Functions (CVPR 2025 · similarity 0.89)
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.