Research map / 3D vision / Face and shape
PatchComplete: Learning Multi-Resolution Patch Priors for 3D Shape Completion on Unseen Categories
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 multi-resolution patch priors instead of only global 32³ priors (ablation) | PatchComplete · ShapeNet novel categories | Inst-CD: 11.97 ±3{e}^{-2} -> 4.23 ±4{e}^{-2} | Helped |
| Use multi-resolution patch priors instead of only 8³ priors (ablation) | PatchComplete · ShapeNet novel categories | Inst-CD: 4.89 ±3{e}^{-2} -> 4.23 ±4{e}^{-2} | Helped |
| Use multi-resolution patch priors instead of only 4³ priors (ablation) | PatchComplete · ShapeNet novel categories | Inst-CD: 4.45 ±1{e}^{-2} -> 4.23 ±4{e}^{-2} | Helped |
| Use cross-attention rather than MLP-predicted input-prior weights (ablation) | PatchComplete · ShapeNet novel categories | Inst-CD: 4.90 -> 4.69 | Helped |
| Use cross-attention rather than MLP-predicted input-prior weights (ablation) | PatchComplete · ScanNet novel categories | Inst-CD: 7.80 -> 7.58 | Helped |
| Train priors end-to-end rather than keep them fixed (ablation) | PatchComplete · ShapeNet novel categories | Inst-CD: 4.31 -> 4.23 | Helped |
| Pre-train on synthetic data before ScanNet training (ablation) | PatchComplete · ScanNet novel categories | Inst-CD: 7.61 -> 7.38 | Helped |
| Increase prior count from 50% to the standard count (ablation) | PatchComplete · ShapeNet novel categories | Inst-CD: 4.41 -> 4.23 | Helped |
| Increase prior count from the standard count to 150% (ablation) | PatchComplete · ShapeNet novel categories | Inst-CD: 4.23 -> 4.22 | No clear effect |
| Combine 4³ and 32³ resolutions rather than use all resolutions (ablation) | PatchComplete · ShapeNet novel categories | Inst-CD: 4.30 -> 4.23 | Helped |
| Combine 4³ and 8³ resolutions rather than use all resolutions (ablation) | PatchComplete · ShapeNet novel categories | Inst-CD: 4.35 -> 4.23 | Helped |
| Concatenate observed input features with the attention result (ablation) | PatchComplete · ShapeNet novel categories | Inst-CD: 4.25 -> 4.23 | Helped |
Closest papers by meaning
- Learning to Reconstruct Shapes from Unseen Classes (NeurIPS 2018 · similarity 0.90)
- AutoSDF: Shape Priors for 3D Completion, Reconstruction and Generation (CVPR 2022 · similarity 0.89)
- Fostering Generalization in Single-View 3D Reconstruction by Learning a Hierarchy of Local and Global Shape Priors (CVPR 2021 · similarity 0.89)
- DinoComplete: 3D Shape Completion with Distilled Semantic Priors and State Space Models (NeurIPS 2026 · similarity 0.89)
- DiffComplete: Diffusion-based Generative 3D Shape Completion (NeurIPS 2023 · similarity 0.89)
- 3D Shape Completion with Multi-View Consistent Inference (AAAI 2020 · similarity 0.89)
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.