Research map / Point clouds / Shape and surface completion
CPFN: Cascaded Primitive Fitting Networks for High-Resolution Point Clouds
Eric-Tuan Lê, Minhyuk Sung, Duygu Ceylan, Radomir Mech, Tamy Boubekeur, Niloy J. Mitra
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 local cascaded SPFN (CPFN, eta=5%) on top of global SPFN | SPFN · TraceParts high-resolution primitive fitting (128k points) | segmentation mIoU (%): 66.29 -> 79.64 | Improved |
| add local cascade | SPFN · TraceParts high-resolution primitive fitting (128k points) | primitive type accuracy (%): 89.50 -> 96.45 | Improved |
| increase patch scale threshold eta from 1% to 5% | CPFN · TraceParts high-resolution primitive fitting (128k points) | segmentation mIoU (%): 70.11 -> 79.64 | Improved |
| add local cascade for smallest primitives (<1% scale) | SPFN · TraceParts high-resolution primitive fitting (128k points) | per-scale mIoU: 44.25 -> 65.74 | Improved |
| add object-level global features l_o (ablation) | CPFN local SPFN · TraceParts high-resolution primitive fitting (128k points) ablation | segmentation mIoU (%): 77.70 -> 78.30 | Improved |
| add patch features l_i^g from global SPFN (ablation) | CPFN local SPFN · TraceParts high-resolution primitive fitting (128k points) ablation | segmentation mIoU (%): 77.70 -> 77.95 | Improved |
| add both l_o and l_i^g (ablation) | CPFN local SPFN · TraceParts high-resolution primitive fitting (128k points) ablation | primitive axis difference (deg): 3.07 -> 1.44 | Improved |
| add global SPFN predictions in merging (ablation) | CPFN · TraceParts high-resolution primitive fitting (128k points) ablation | segmentation mIoU (%): 74.13 -> 79.64 | Improved |
| add patch selection network (ablation) | CPFN · TraceParts high-resolution primitive fitting (128k points) ablation | segmentation mIoU (%): 78.44 -> 79.64 | Improved |
| add patch selection network (ablation) | CPFN · TraceParts high-resolution primitive fitting (128k points) ablation | P coverage eps=0.01 (%): 74.42 -> 88.73 | Improved |
| use ground truth patch selection instead of learned one (ablation) | CPFN · TraceParts high-resolution primitive fitting (128k points) ablation | segmentation mIoU (%): 79.64 -> 80.94 | No clear effect |
| use CPFN instead of SPFN at 16k points (ablation) | SPFN · TraceParts (16k points) | segmentation mIoU (%): 65.86 -> 74.77 | Improved |
Closest papers by meaning
- Supervised Fitting of Geometric Primitives to 3D Point Clouds (CVPR 2019 · similarity 0.89)
- Finding Good Configurations of Planar Primitives in Unorganized Point Clouds (CVPR 2022 · similarity 0.88)
- PrimitiveUDF: Primitive-Based Unsigned Distance Fields for Surface Reconstruction from Point Clouds (ECCV 2026 · similarity 0.87)
- DeepLA-Net: Very Deep Local Aggregation Networks for Point Cloud Analysis (CVPR 2025 · similarity 0.85)
- LinNet: Linear Network for Efficient Point Cloud Representation Learning (NeurIPS 2024 · similarity 0.85)
- Point Cloud Pre-training with Diffusion Models (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.
