Research map / Point clouds / Shape and surface completion
Learning Representations and Generative Models for 3D Point Clouds
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Generate samples in a learned AE latent space rather than directly from raw point clouds | GAN point-cloud generation · ShapeNet chair test | COV-EMD: 19.0 -> 57.1 | Helped |
| Use an EMD-trained AE instead of a CD-trained AE for the latent GAN (ablation) | l-GAN · ShapeNet chair test | COV-EMD: 32.2 -> 57.1 | Helped |
| Use an EMD-trained AE instead of a CD-trained AE for the latent GAN (ablation) | l-GAN · ShapeNet chair test | MMD-EMD: 0.079 -> 0.069 | Helped |
| Replace the standard latent GAN objective with a Wasserstein objective and gradient penalty (ablation) | l-GAN (AE-EMD) · ShapeNet chair test | COV-EMD: 57.1 -> 66.9 | Helped |
| Fit a GMM instead of using a latent WGAN (ablation) | AE-EMD latent generator · ShapeNet chair test | JSD: 0.022 -> 0.020 | Helped |
| Use a full covariance matrix rather than a diagonal covariance matrix (ablation) | GMM in AE-EMD latent space · ShapeNet chair | JSD | Helped |
| Use a bottleneck size of 128 rather than other tested sizes (ablation) | AE · ShapeNet chair test | generalization error | Helped |
| Add dropout layers (ablation) | AE · ShapeNet chair | reconstruction quality | Hurt |
| Use batch normalization on the encoder only (ablation) | AE · Single-class data | generalization error | Helped |
| Train the completion network with CD rather than EMD loss (ablation) | Abstractor-Predictor · ShapeNet Chair test | coverage: 82.6 -> 77.5 | Hurt |
| Train the completion network with CD rather than EMD loss (ablation) | Abstractor-Predictor · ShapeNet Airplane test | accuracy: 94.7 -> 96.9 | Helped |
| Train the completion network with CD rather than EMD loss (ablation) | Abstractor-Predictor · ShapeNet Table test | coverage: 83.0 -> 75.2 | Hurt |
Closest papers by meaning
- Learning Progressive Point Embeddings for 3D Point Cloud Generation (CVPR 2021 · similarity 0.91)
- 3D Shape Generation With Grid-Based Implicit Functions (CVPR 2021 · similarity 0.89)
- Self-Supervised Learning on 3D Point Clouds by Learning Discrete Generative Models (CVPR 2021 · similarity 0.89)
- Learning Localized Generative Models for 3D Point Clouds via Graph Convolution (ICLR 2019 · similarity 0.89)
- Generative PointNet: Deep Energy-Based Learning on Unordered Point Sets for 3D Generation, Reconstruction and Classification (CVPR 2021 · similarity 0.88)
- PointGMM: A Neural GMM Network for Point Clouds (CVPR 2020 · 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.