Research map / 3D vision / Face and shape
Deep Single-View 3D Object Reconstruction with Visual Hull Embedding
Hanqing Wang, Jiaolong Yang, Wei Liang, Xin Tong
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 single-view visual-hull refinement to the coarse V-Net shape | V-Net reconstruction · ShapeNet mean IoU | IoU: 0.631 -> 0.680 | Improved |
| add single-view visual-hull refinement to the coarse V-Net shape | V-Net reconstruction · PASCAL 3D+ car, airplane, chair, and couch | IoU: car: 0.625 -> 0.653; airplane: 0.647 -> 0.690; chair: 0.341 -> 0.341; couch: 0.633 -> 0.664 | Mixed |
| remove the probabilistic visual hull input (ablation) | R-Net refinement · ShapeNet mean IoU | IoU: 0.680 -> 0.636 | Hurt |
| replace the estimated visual hull with a ground-truth visual hull (ablation) | R-Net refinement · ShapeNet mean IoU | IoU: 0.631 -> 0.726 | Improved |
| add the two probability maps V⊙(1-H) and H⊙(1-V) to R-Net input (ablation) | R-Net refinement · ShapeNet mean IoU | IoU: 0.675 -> 0.680 | Improved |
| add end-to-end finetuning (ablation) | the full reconstruction network · ShapeNet mean IoU | IoU: 0.658 -> 0.680 | Improved |
| use natural image crop backgrounds instead of clean backgrounds | the proposed reconstruction method · ShapeNet mean IoU | IoU: 0.680 -> 0.674 | Hurt |
| use the proposed visual-hull refinement method instead of PointOutNet | ShapeNet reconstruction · ShapeNet car, airplane, chair, and couch | IoU: car: 0.831 -> 0.839; airplane: 0.601 -> 0.631; chair: 0.544 -> 0.552; couch: 0.708 -> 0.698 | Mixed |
Closest papers by meaning
- Image-Based 3D Object Reconstruction: State-of-the-Art and Trends in the Deep Learning Era (TPAMI 2019 · similarity 0.90)
- Multiview Aggregation for Learning Category-Specific Shape Reconstruction (NeurIPS 2019 · similarity 0.88)
- Single-View 3D Object Reconstruction From Shape Priors in Memory (CVPR 2021 · similarity 0.88)
- Efficient Dense Point Cloud Object Reconstruction using Deformation Vector Fields (ECCV 2018 · similarity 0.88)
- Learning Shape Priors for Single-View 3D Completion and Reconstruction (ECCV 2018 · similarity 0.87)
- What Do Single-View 3D Reconstruction Networks Learn? (CVPR 2019 · similarity 0.87)
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.
