Research map / Point clouds / Shape and surface completion
Supervised Fitting of Geometric Primitives to 3D Point Clouds
Lingxiao Li, Minhyuk Sung, Anastasia Dubrovina, Li Yi, Leonidas J. Guibas
Check 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 |
|---|---|---|---|
| Predict per-point properties and estimate primitives differentiably instead of directly regressing primitive parameters | SPFN vs. DPPN · ANSI mechanical component test set | Primitive surface coverage (ε=0.01): 46.99 -> 86.63 | Improved |
| Use SPFN-predicted segmentation with Efficient RANSAC | Efficient RANSAC · ANSI mechanical component test set | Primitive surface coverage (ε=0.01): 56.95 -> 74.32 | Improved |
| Add predicted primitive types and normals to segmented Efficient RANSAC | Efficient RANSAC with SPFN segmentation · ANSI mechanical component test set | Primitive surface coverage (ε=0.01): 74.32 -> 65.31 -> 61.94 | Hurt |
| Add segmentation loss (ablation) | SPFN · ANSI mechanical component test set | Primitive surface coverage (ε=0.01): 50.04 -> 86.63 | Improved |
| Predict point normals and train with normal loss rather than supplying jet-fitting normals from 64k points (ablation) | SPFN · ANSI mechanical component test set | Primitive surface coverage (ε=0.01): 76.47 -> 86.63 | Improved |
| Add fitting residual loss (ablation) | SPFN · ANSI mechanical component test set | Primitive surface coverage (ε=0.01): 79.81 -> 86.63 | Improved |
| Add fitting residual loss (ablation) | SPFN · ANSI mechanical component test set | Segmentation Mean IoU: 72.70 -> 77.14 | Improved |
| Add primitive axis loss (ablation) | SPFN · ANSI mechanical component test set | Primitive axis difference (°): 6.27 -> 1.51 | Improved |
| Add primitive axis loss (ablation) | SPFN · ANSI mechanical component test set | Primitive surface coverage (ε=0.01): 80.80 -> 86.63 | Improved |
| Use ground-truth rather than predicted primitive types in the fitting residual loss (ablation) | SPFN · ANSI mechanical component test set | Primitive surface coverage (ε=0.01): 85.25 -> 86.63 | Improved |
| Increase test input resolution from 8k to 64k points | SPFN trained with 8k points · ANSI mechanical component test set | Primitive surface coverage (ε=0.01): 86.63 -> 87.03 | Improved |
| Add 10% uniformly sampled outliers during training and testing | SPFN · ANSI mechanical component test set with outliers | Primitive surface coverage (ε=0.01): 86.63 -> 82.57 | Hurt |
Closest papers by meaning
- CPFN: Cascaded Primitive Fitting Networks for High-Resolution Point Clouds (ICCV 2021 · similarity 0.89)
- DeepFit: 3D Surface Fitting via Neural Network Weighted Least Squares (ECCV 2020 · similarity 0.87)
- SpelsNet: Surface Primitive Elements Segmentation by B-Rep Graph Structure Supervision (NeurIPS 2024 · similarity 0.85)
- A Lightweight UDF Learning Framework for 3D Reconstruction Based on Local Shape Functions (CVPR 2025 · similarity 0.85)
- SPU-PMD: Self-Supervised Point Cloud Upsampling via Progressive Mesh Deformation (CVPR 2024 · similarity 0.85)
- X-3D: Explicit 3D Structure Modeling for Point Cloud Recognition (CVPR 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.
