Research map / Point clouds / Shape and surface completion
Learning Local Pattern Modularization for Point Cloud Reconstruction from Unseen Classes
Chao Chen, Yu-Shen Liu, Zhizhong Han
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 prior (pattern modularization plus customization) over initial shape decoder only (ablation) | Ours · ShapeNet Bench (seen, 2048 pts) | L1-CD: 0.054 -> 0.032 | Improved |
| add pattern learner and pattern modularization (ablation) | Ours · ShapeNet Bench | L1-CD: 0.049 -> 0.032 | Improved |
| add modularization customizer shift from the image (ablation) | Ours · ShapeNet Bench | L1-CD: 0.041 -> 0.032 | Improved |
| add region loss L_Region instead of repeating L_Shape (ablation) | Ours · ShapeNet Bench | L1-CD: 0.038 -> 0.032 | Improved |
| add weak global shape loss L_Shape (ablation) | Ours · ShapeNet Bench | L1-CD: 0.048 -> 0.032 | Improved |
| sample pattern points in 3D voxel instead of on a 2D plane (ablation) | pattern learner · ShapeNet Bench | L1-CD: 0.034 -> 0.032 | Improved |
| vary region number M from 8 to 1 or 27 (ablation) | Ours · ShapeNet Bench | L1-CD: 0.032 -> 0.035 (M=1), 0.049 (M=27) | Hurt |
| increase pattern number from 8 to 16 (ablation) | Ours · ShapeNet Bench | L1-CD: 0.032 -> 0.035 | Hurt |
| use local prior instead of global prior for unseen class reconstruction | Ours vs DRWR · ShapeNet 4 unseen classes (1024 pts) | L1-CD: 0.0715 -> 0.0540 | Improved |
| train on 3 seen classes while competitors use their full training settings | Ours · ShapeNet seen (5-class protocol) | L1-CD: 0.0536 -> 0.0527 | Improved |
| use ours versus SDT on Chair, Sofa and Table classes | Ours · ShapeNet seen classes (2048 pts) | L1-CD: Chair 0.036 -> 0.040; Table 0.033 -> 0.044 | Mixed |
Closest papers by meaning
- Fostering Generalization in Single-View 3D Reconstruction by Learning a Hierarchy of Local and Global Shape Priors (CVPR 2021 · similarity 0.90)
- Surface Reconstruction From Point Clouds by Learning Predictive Context Priors (CVPR 2022 · similarity 0.90)
- A Survey and Benchmark of Automatic Surface Reconstruction From Point Clouds (TPAMI 2024 · similarity 0.89)
- Learning to Reconstruct Shapes from Unseen Classes (NeurIPS 2018 · similarity 0.88)
- A Lightweight UDF Learning Framework for 3D Reconstruction Based on Local Shape Functions (CVPR 2025 · similarity 0.87)
- Learning Partonomic 3D Reconstruction from Image Collections (CVPR 2025 · similarity 0.87)
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.
