Research map / 3D vision / Face and shape
High-Resolution Shape Completion Using Deep Neural Networks for Global Structure and Local Geometry Inference
Xiaoguang Han, Zhen Li, Haibin Huang, Evangelos Kalogerakis, Yizhou Yu
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 |
|---|---|---|---|
| add AUC loss (ablation) | global structure network · Chair ablation (global network) | F1 of inside voxels: 0.904 -> 0.926 | Improved |
| add projected depth images (2D branch) (ablation) | global structure network · Chair ablation (global network) | F1 of inside voxels: 0.877 -> 0.926 | Improved |
| add BLSTM context modeling (ablation) | global structure network · Chair ablation (global network) | F1 of inside voxels: 0.896 -> 0.926 | Improved |
| add BSurf channel to CSDF input (ablation) | global structure network · Chair ablation (global network) | F1 of inside voxels: 0.90 -> 0.926 | Improved |
| add CSDF channel to BSurf-only input (ablation) | global structure network · Chair ablation (global network) | F1 of inside voxels: 0.836 -> 0.926 | Improved |
| use 1x1x1 conv head instead of encoder-decoder (ablation) | global structure network · Chair ablation (global network) | F1 of inside voxels: 0.818 -> 0.926 | Improved |
| add global guidance from global network (ablation) | local refinement network · Chair ablation (local network) | accuracy: 0.912 -> 0.961 | Improved |
| local refinement at 256^3 instead of 32^3 global network alone | whole network · Chair completion | completeness: 70.1% -> 97.25% | Improved |
| local refinement at 256^3 instead of 32^3 global network alone | whole network · Chair completion | normalized distance: 0.0108 -> 0.00398 | Improved |
| patch-based learned completion instead of Poisson reconstruction | whole network · Guitar completion | normalized distance: 0.148 -> 0.00248 | Improved |
| learned completion instead of Poisson reconstruction | whole network · Car completion | completeness: 82.18% -> 95.88% | Improved |
| larger missing regions (one side missing, 55.68% input completeness) | whole network · Chair, large missing regions | F1 of global network: 0.926 -> 0.895 | Mixed |
Closest papers by meaning
- Shape Completion Using 3D-Encoder-Predictor CNNs and Shape Synthesis (CVPR 2017 · similarity 0.91)
- Learning Unsupervised Hierarchical Part Decomposition of 3D Objects From a Single RGB Image (CVPR 2020 · similarity 0.89)
- Deep Local Shapes: Learning Local SDF Priors for Detailed 3D Reconstruction (ECCV 2020 · similarity 0.89)
- 3D Shape Completion with Multi-View Consistent Inference (AAAI 2020 · similarity 0.89)
- Learning Shape Priors for Single-View 3D Completion and Reconstruction (ECCV 2018 · similarity 0.89)
- Learning Generalizable Shape Completion with SIM(3) Equivariance (NeurIPS 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.
