Research map / 3D vision / Face and shape
Fostering Generalization in Single-View 3D Reconstruction by Learning a Hierarchy of Local and Global Shape Priors
Check 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 |
|---|---|---|---|
| Replace a global ONet prior with a hierarchy of local and global priors | Single-view 3D reconstruction · ShapeNet multi-class training, mean unseen categories | F-score: 29.3 -> 48.2 | Helped |
| Replace a global ONet prior with a hierarchy of local and global priors | Single-view 3D reconstruction · ShapeNet multi-class training, Composition | F-score: 18.3 -> 42.4 | Helped |
| Use signed-distance prediction for the global level in HPN (ablation) | HPN · ShapeNet multi-class training, mean unseen categories | F-score: 42.9 -> 48.2 | Helped |
| Use signed-distance prediction for the global level in HPN (ablation) | HPN · ShapeNet lamp-only training, mean unseen categories | F-score: 47.1 -> 46.1 | Hurt |
| Add Local@64 to a global reconstruction network (ablation) | Global@256 (ONet) · ShapeNet unseen categories, chair-only training, full shape | F-score: 23.3 -> 42.0 | Helped |
| Fuse Global@256 and Local@64 (ablation) | Global@256 (ONet) · ShapeNet unseen categories, chair-only training, full shape | F-score: 23.3 -> 38.3 | Helped |
| Add Local@32 to the Global@256 and Local@64 fusion (ablation) | HPN@(256+64) · ShapeNet unseen categories, chair-only training, full shape | F-score: 38.3 -> 39.7 | Helped |
| Add Local@128 and Local@16 to HPN@(256+64+32) (ablation) | HPN@(256+64+32) · ShapeNet unseen categories, chair-only training, full shape | Chamfer distance: 4.7 -> 4.4 | Helped |
| Use the five-level hierarchy instead of HPN@(256+64+32) (ablation) | HPN · ShapeNet unseen categories, chair-only training, full shape | F-score: 39.7 -> 42.0 | Helped |
| Use very small local patches alone rather than combining locality levels (ablation) | Local reconstruction networks · ShapeNet unseen categories, chair-only training, invisible regions | Chamfer distance: Local@16: 8.7; Local@32: 7.4 | Hurt |
| Train the local network with 1% of the original data rather than the full training data (ablation) | Local@64 · ShapeNet multi-class, unseen categories | F-score: null | No clear effect |
| Use a local prior instead of a global prior (ablation) | Single-view reconstruction · Visible regions of unseen shapes | F-score: 28.6 -> 67.3 | Helped |
Closest papers by meaning
- Learning to Reconstruct Shapes from Unseen Classes (NeurIPS 2018 · similarity 0.92)
- 3D Reconstruction with Generalizable Neural Fields using Scene Priors (ICLR 2024 · similarity 0.88)
- MonoInstance: Enhancing Monocular Priors via Multi-view Instance Alignment for Neural Rendering and Reconstruction (CVPR 2025 · similarity 0.88)
- Leveraging Global Stereo Consistency for Category-Level Shape and 6D Pose Estimation from Stereo Images (CVPR 2025 · similarity 0.88)
- Zero-Shot Scene Reconstruction from Single Images with Deep Prior Assembly (NeurIPS 2024 · similarity 0.88)
- 3D-Aware Hypothesis & Verification for Generalizable Relative Object Pose Estimation (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.