Research map / Point clouds / Shape and surface completion
Discrete Point Flow Networks for Efficient Point Cloud Generation
Roman Klokov, Edmond Boyer, Jakob Verbeek
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 |
|---|---|---|---|
| replace continuous flows (PointFlow) with discrete affine coupling flows (DPF-Net) | PointFlow-style model · ShapeNet generation | total training time (days): 80 -> 1.1 | Improved |
| replace continuous flows with discrete affine coupling flows | PointFlow-style model · ShapeNet generation | generation time (ms/sample): 150 -> 4 | Improved |
| replace continuous flows with discrete coupling flows | PointFlow-style model · ShapeNet generation | training memory (Mb/sample): 470 -> 370 | Improved |
| train DPF-Net on normalized rather than non-normalized meshes (ablation) | DPF-Net · ShapeNet autoencoding | CD (x10^4): 6.85 -> 6.17 | Improved |
| train DPF-Net on normalized rather than non-normalized meshes (ablation) | DPF-Net · ShapeNet autoencoding | EMD (x10^2): 5.06 -> 4.37 | Improved |
| DPF-Net instead of PointFlow under equal training time | autoencoder · ShapeNet autoencoding | CD (x10^4): 10.22 -> 6.85 | Improved |
| DPF-Net versus PointFlow on car class | generative model · ShapeNet car generation | 1-NNA CD (%): 70.1 -> 70.3 | Hurt |
| DPF-Net versus PointFlow on airplane | generative model · ShapeNet airplane generation | JSD (x10^2): 1.42 -> 0.94 | Improved |
| DPF-Net versus l-GAN-CD on EMD-based 1-NNA | generative model · ShapeNet airplane generation | 1-NNA EMD (%): 92.1 -> 67.0 | Improved |
| DPF-Net versus AtlasNet in single-view reconstruction | single-view reconstruction · ShapeNet SVR | CD (x10^3): 5.34 -> 5.51 | Mixed |
| DPF-Net versus voxel-based PRN | single-view reconstruction · ShapeNet SVR | F1 (%): 53.1 -> 52.4 | No clear effect |
Closest papers by meaning
- Learning Representations and Generative Models for 3D Point Clouds (ICML 2018 · similarity 0.92)
- Learning Progressive Point Embeddings for 3D Point Cloud Generation (CVPR 2021 · similarity 0.91)
- A Point Set Generation Network for 3D Object Reconstruction From a Single Image (CVPR 2017 · similarity 0.90)
- 3DILG: Irregular Latent Grids for 3D Generative Modeling (NeurIPS 2022 · similarity 0.90)
- GaussianAnything: Interactive Point Cloud Flow Matching for 3D Generation (ICLR 2025 · similarity 0.89)
- Not-So-Optimal Transport Flows for 3D Point Cloud Generation (ICLR 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.
