Research map / Point clouds / Shape and surface completion
Unsupervised Learning of Fine Structure Generation for 3D Point Clouds by 2D Projections Matching
Chao Chen, Zhizhong Han, Yu-Shen Liu, Matthias Zwicker
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 the second directional distance term to a first-term-only loss (ablation) | 2D projection matching loss · ShapeNet Chair (32^2 images, 2000 points) | Chamfer distance x100: 24.47 -> 4.16 | Improved |
| add the first directional distance term to a second-term-only loss (ablation) | 2D projection matching loss · ShapeNet Chair (32^2 images, 2000 points) | Chamfer distance x100: 8.49 -> 4.16 | Improved |
| use 5 nearest neighbors in first term (ablation) | 2D projection matching loss · ShapeNet Chair (32^2 images, 2000 points) | Chamfer distance x100: 4.16 -> 4.25 | No clear effect |
| use 5 nearest neighbors in second term (ablation) | 2D projection matching loss · ShapeNet Chair (32^2 images, 2000 points) | Chamfer distance x100: 4.16 -> 4.32 | No clear effect |
| use 5 nearest neighbors in both terms (ablation) | 2D projection matching loss · ShapeNet Chair (32^2 images, 2000 points) | Chamfer distance x100: 4.16 -> 4.29 | No clear effect |
| increase irregular supervision points from 1000 to 5000 (ablation) | 2D projection matching · ShapeNet Chair (32^2 images, 2000 points) | Chamfer distance x100: 4.29 -> 4.16 | Improved |
| increase irregular supervision points beyond 5000 to 9000 (ablation) | 2D projection matching · ShapeNet Chair (32^2 images, 2000 points) | Chamfer distance x100: 4.16 -> 4.19 | No clear effect |
| use structure adaptive sampling instead of Poisson-disk sampling (ablation) | 2D projection matching · ShapeNet Chair (32^2 images, 2000 points) | Chamfer distance x100: 4.35 -> 4.16 | Improved |
| use structure adaptive sampling instead of random sampling (ablation) | 2D projection matching · ShapeNet Chair (32^2 images, 2000 points) | Chamfer distance x100: 4.23 -> 4.16 | Improved |
| use structure adaptive sampling instead of pixel sampling (ablation) | 2D projection matching · ShapeNet Chair (32^2 images, 2000 points) | Chamfer distance x100: 4.20 -> 4.16 | Improved |
| use structure adaptive sampling instead of dynamic random sampling (ablation) | 2D projection matching · ShapeNet Chair (32^2 images, 2000 points) | Chamfer distance x100: 4.36 -> 4.16 | Improved |
| replace per-pixel difference supervision (DRWR) with 2D projection matching | point cloud generation · ShapeNet Chair 16000 points, 128^2 to 32^2 images | CD degradation x100: 0.46 -> 0.31 | Improved |
Closest papers by meaning
- UniPre3D: Unified Pre-training of 3D Point Cloud Models with Cross-Modal Gaussian Splatting (CVPR 2025 · similarity 0.89)
- From Image Collections to Point Clouds With Self-Supervised Shape and Pose Networks (CVPR 2020 · similarity 0.89)
- CAPNet: Continuous Approximation Projection for 3D Point Cloud Reconstruction Using 2D Supervision (AAAI 2019 · similarity 0.88)
- A Point Set Generation Network for 3D Object Reconstruction From a Single Image (CVPR 2017 · similarity 0.88)
- PointGLR: Unsupervised Structural Representation Learning of 3D Point Clouds (TPAMI 2022 · similarity 0.88)
- Multi-View Representation is What You Need for Point-Cloud Pre-Training (ICLR 2024 · 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.
