Research map / Point clouds / Shape and surface completion
Unsupervised Learning of Shape and Pose with Differentiable 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 |
|---|---|---|---|
| Use point clouds rather than voxels | Ours, known-pose shape prediction · ShapeNet, resolution 32 | Mean Chamfer distance ×100: 5.01 -> 4.61 | Helped |
| Use point clouds rather than voxels | Ours, known-pose shape prediction · ShapeNet, resolution 64 | Mean Chamfer distance ×100: 4.39 -> 3.55 | Helped |
| Train with higher-resolution supervision | Ours, known-pose shape prediction · ShapeNet mean across airplane, car, and chair | Mean Chamfer distance ×100: 4.61 -> 3.55 -> 2.96 | Helped |
| Use the ensemble of pose predictors rather than one predictor | Ours, joint shape and pose estimation · ShapeNet, mean | Chamfer distance: 5.38 -> 3.89 | Helped |
| Use the ensemble of pose predictors rather than one predictor | Ours, joint shape and pose estimation · ShapeNet, mean | Pose accuracy: 0.40 -> 0.82 | Helped |
| Use the ensemble of pose predictors rather than one predictor | Ours, joint shape and pose estimation · ShapeNet, mean | Median pose error: 58.1 -> 7.1 | Helped |
| Add point dropout during training (ablation) | Full point-cloud shape model · ShapeNet chairs, 64×64, known pose | Chamfer distance ×100: 4.59 -> 4.03 | Helped |
| Use a scheduled point size rather than fixed σ=1.6 (ablation) | Full point-cloud shape model · ShapeNet chairs, 64×64, known pose | Chamfer distance ×100: 4.89 -> 4.03 | Helped |
| Learn point scale rather than fixing it (ablation) | Full point-cloud shape model · ShapeNet chairs, 64×64, known pose | Chamfer distance ×100: 4.27 -> 4.03 | Helped |
| Use 8000 rather than 4000 points (ablation) | Full point-cloud shape model · ShapeNet chairs, 64×64, known pose | Chamfer distance ×100: 4.28 -> 4.03 | Helped |
| Use 8000 rather than 2000 points (ablation) | Full point-cloud shape model · ShapeNet chairs, 64×64, known pose | Chamfer distance ×100: 4.70 -> 4.03 | Helped |
| Use 8000 rather than 1000 points (ablation) | Full point-cloud shape model · ShapeNet chairs, 64×64, known pose | Chamfer distance ×100: 5.10 -> 4.03 | Helped |
Closest papers by meaning
- Shape-Pose Ambiguity in Learning 3D Reconstruction from Images (AAAI 2021 · similarity 0.90)
- From Image Collections to Point Clouds With Self-Supervised Shape and Pose Networks (CVPR 2020 · similarity 0.89)
- Unsupervised Learning of Category-Level 3D Pose from Object-Centric Videos (CVPR 2024 · similarity 0.88)
- Shelf-Supervised Mesh Prediction in the Wild (CVPR 2021 · similarity 0.88)
- Learning Shape-Independent Transformation via Spherical Representations for Category-Level Object Pose Estimation (ICLR 2025 · similarity 0.88)
- Rectified Point Flow: Generic Point Cloud Pose Estimation (NeurIPS 2025 · similarity 0.86)
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.