Research map / Point clouds / Shape and surface completion
Neural-Pull: Learning Signed Distance Function from Point clouds by Learning to Pull Space onto Surface
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 SDFs by pulling sampled query points to their nearest point-cloud surface points using predicted signed distances and gradients | Neural-Pull · ABC | L2-CD (×100): 0.51 -> 0.48 | Helped |
| Learn SDFs by pulling sampled query points to their nearest point-cloud surface points using predicted signed distances and gradients | Neural-Pull · FAMOUS | L2-CD (×100): 1.41 -> 0.22 | Helped |
| Use Neural-Pull for surface reconstruction | FAMOUS ShapeNet subset · FAMOUS ShapeNet subset | Normal consistency: 0.962 -> 0.939 | Mixed |
| Use Neural-Pull for single-image reconstruction | ShapeNet single-image reconstruction · ShapeNet single-image, mean | L1-CD: 0.175 -> 0.021 | Helped |
| Add geometric network initialization (ablation) | Neural-Pull · FAMOUS | L2-CD (×100): 0.35 -> 0.22 | Helped |
| Sample query locations near the surface rather than uniformly in the entire 3D space (ablation) | Neural-Pull · FAMOUS | L2-CD (×100): 0.80 -> 0.22 | Helped |
| Add the IGR gradient-norm constraint (ablation) | Neural-Pull · FAMOUS | L2-CD (×100): 0.22 -> 1.15 | Hurt |
| Increase query-location count from 1×10^6 to 10×10^6 (ablation) | Neural-Pull · FAMOUS | L2-CD (×100): 0.434 -> 0.221 | Helped |
| Increase query-location count from 5×10^6 to 10×10^6 (ablation) | Neural-Pull · FAMOUS | training time per epoch: 530 -> 1020 | Hurt |
| Increase ground-truth point-cloud resolution (ablation) | Neural-Pull · FAMOUS | L2-CD (×100): 0.293 -> 0.213 | Helped |
| Use query locations sampled at the standard range instead of a too-small range (ablation) | Neural-Pull · FAMOUS | L2-CD (×100): 0.348 -> 0.223 | Helped |
| Use query locations sampled at the standard range instead of a too-large range (ablation) | Neural-Pull · FAMOUS | L2-CD (×100): 0.271 -> 0.223 | Helped |
Closest papers by meaning
- MultiPull: Detailing Signed Distance Functions by Pulling Multi-Level Queries at Multi-Step (NeurIPS 2024 · similarity 0.90)
- Small Steps and Level Sets: Fitting Neural Surface Models with Point Guidance (CVPR 2024 · similarity 0.89)
- A Lightweight UDF Learning Framework for 3D Reconstruction Based on Local Shape Functions (CVPR 2025 · similarity 0.89)
- Neural Signed Distance Function Inference through Splatting 3D Gaussians Pulled on Zero-Level Set (NeurIPS 2024 · similarity 0.89)
- Inferring Neural Signed Distance Functions by Overfitting on Single Noisy Point Clouds through Finetuning Data-Driven based Priors (NeurIPS 2024 · similarity 0.88)
- Learning Bijective Surface Parameterization for Inferring Signed Distance Functions from Sparse Point Clouds with Grid Deformation (CVPR 2025 · similarity 0.88)
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.