Research map / 3D vision / Face and shape
Learning Single-Image 3D Reconstruction by Generative Modelling of Shape, Pose and Shading
Paul D. Henderson, Vittorio Ferrari
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 |
|---|---|---|---|
| use shading loss instead of silhouette-only loss (ablation) | ours (coloured light) · ShapeNet car | IoU: 0.46 -> 0.77 | Improved |
| use shading loss instead of silhouette-only loss (ablation) | ours (coloured light) · ShapeNet aeroplane | IoU: 0.20 -> 0.55 | Improved |
| use shading loss instead of silhouette-only loss (ablation) | ours (coloured light) · ShapeNet sofa | IoU: 0.27 -> 0.59 | Improved |
| use three coloured lights instead of one white light (ablation) | ours · ShapeNet car | IoU: 0.58 -> 0.77 | Improved |
| use three coloured lights instead of one white light (ablation) | ours · ShapeNet chair | IoU: 0.31 -> 0.46 | Improved |
| add ground-truth pose supervision (ablation) | ours · ShapeNet car | IoU: 0.77 -> 0.82 | Improved |
| train with three views per instance, test with one (ablation) | ours · ShapeNet car | IoU: 0.77 -> 0.82 | Improved |
| test with three views as well (ablation) | ours · ShapeNet chair | IoU: 0.46 -> 0.53 | Improved |
| vary lighting angle across images (unknown to model) (ablation) | ours (coloured light) · ShapeNet car | IoU: 0.77 -> 0.60 | Hurt |
| vary lighting angle across images (unknown to model) (ablation) | ours (coloured light) · ShapeNet sofa | IoU: 0.59 -> 0.69 | Improved |
| use full-block instead of subdivision parameterisation (ablation) | ours · ShapeNet chair | IoU: 0.39 -> 0.46 | Improved |
| use subdivision instead of ortho-block parameterisation (ablation) | ours · ShapeNet aeroplane | IoU: 0.30 -> 0.55 | Improved |
Closest papers by meaning
- Learning to Reconstruct Shapes from Unseen Classes (NeurIPS 2018 · similarity 0.89)
- Self-supervised Single-view 3D Reconstruction via Semantic Consistency (ECCV 2020 · similarity 0.89)
- Shape-Pose Ambiguity in Learning 3D Reconstruction from Images (AAAI 2021 · similarity 0.89)
- Learning Shape Priors for Single-View 3D Completion and Reconstruction (ECCV 2018 · similarity 0.89)
- High-Fidelity 3D Textured Shapes Generation by Sparse Encoding and Adversarial Decoding (ECCV 2024 · similarity 0.89)
- Learning Partonomic 3D Reconstruction from Image Collections (CVPR 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.
