Research map / 3D vision / Face and shape
Synthesizing 3D Shapes via Modeling Multi-View Depth Maps and Silhouettes With Deep Generative Networks
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 |
|---|---|---|---|
| Use all 20 views rather than only one input view (ablation) | ShapeNet reconstruction with AllVPNet vs SingleVPNet · ShapeNet test set | IoU: 70.7 -> 84.0 | Helped |
| Use all 20 views rather than dropping 15 to 18 input views (ablation) | ShapeNet reconstruction with AllVPNet vs DropoutNet · ShapeNet test set | IoU: 78.5 -> 84.0 | Helped |
| Use depth maps rather than silhouettes (ablation) | AllVPNet unconditional reconstruction · ShapeNet test set | IoU: 83.6 -> 84.0 | Helped |
| Condition on class labels (ablation) | AllVPNet depth reconstruction · ShapeNet test set | IoU: 84.0 -> 83.9 | No clear effect |
| Use one randomly selected input view rather than all views (ablation) | ShapeNet classification with SingleVPNet vs AllVPNet · ModelNet40 test set | accuracy: 82.1% ± 0.1 -> 65.3% ± 0.3 | Hurt |
| Randomly zero out 15 to 18 of the input views (ablation) | ShapeNet classification with DropoutNet vs AllVPNet · ModelNet40 test set | accuracy: 82.1% ± 0.1 -> 74.2% ± 0.2 | Hurt |
| Use depth maps rather than silhouettes (ablation) | SingleVPNet conditional reconstruction · ShapeNet test set | depth reconstruction error (ℓ1): 0.033 -> 0.029 | Helped |
| Use a single input view rather than all views (ablation) | ShapeNet conditional classification · ShapeNet test set | accuracy: 89.1 ± 0.1 -> 82.7 ± 0.1 | Hurt |
| Train with silhouettes rather than depth maps (ablation) | AllVPNet conditional classification · ShapeNet test set | accuracy: 89.1 ± 0.1 -> 87.6 ± 0.1 | Hurt |
Closest papers by meaning
- Visual Object Networks: Image Generation with Disentangled 3D Representations (NeurIPS 2018 · similarity 0.90)
- Multi-View Silhouette and Depth Decomposition for High Resolution 3D Object Representation (NeurIPS 2018 · similarity 0.90)
- SurfNet: Generating 3D Shape Surfaces Using Deep Residual Networks (CVPR 2017 · similarity 0.89)
- Synthesizing 3D Shapes From Silhouette Image Collections Using Multi-Projection Generative Adversarial Networks (CVPR 2019 · similarity 0.89)
- Pixels, Voxels, and Views: A Study of Shape Representations for Single View 3D Object Shape Prediction (CVPR 2018 · similarity 0.89)
- Make-A-Shape: a Ten-Million-scale 3D Shape Model (ICML 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.