Research map / 3D vision / Face and shape
Shape Reconstruction Using Differentiable Projections and Deep Priors
Matheus Gadelha, Rui Wang, Subhransu Maji
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 |
|---|---|---|---|
| replace space carving with deep shape prior and view optimization | Silhouette reconstruction (8 views, perturbed viewpoints) · 12 shapes mean | IoU: 0.55 -> 0.72 | Improved |
| use deep shape prior even with unperturbed viewpoints for the baseline | Silhouette reconstruction · 12 shapes mean | IoU: 0.61 -> 0.72 | Improved |
| perturb viewpoints for space carving (ablation) | Space carving · 12 shapes mean | IoU: 0.61 -> 0.55 | Hurt |
| use deep shape prior instead of space carving on plane | Silhouette reconstruction · plane | IoU: 0.35 vs 0.49 | Hurt |
| use deep shape prior instead of space carving on lamp | Silhouette reconstruction · lamp | IoU: 0.62 vs 0.26 | Improved |
| use deep image prior instead of iterative BM3D | Tomography (n=30, sigma=1) image 1 · Shepp-Logan-type image | SSIM: 0.701 -> 0.967 | Improved |
| use deep image prior instead of iterative BM3D | Tomography (n=30, sigma=1) image 1 · Shepp-Logan-type image | PSNR: 26.0 -> 31.7 | Improved |
| use deep image prior instead of iterative BM3D | Tomography (n=30, sigma=1) image 2 · BrainWeb slice | PSNR: 26.2 -> 26.8 | Improved |
| use deep image prior instead of iterative BM3D | Tomography (n=30, sigma=1) image 3 · BrainWeb slice | SSIM: 0.718 -> 0.834 | Improved |
| add TV prior to FBP reconstruction | Tomography (n=30, sigma=1) · image 1 | PSNR: 15.3 -> 18.6 | Improved |
| increase number of views for depth reconstruction (4 to 32) at high noise (ablation) | Depth-image reconstruction · sigma=0.1 | reconstruction quality (qualitative) | Improved |
Closest papers by meaning
- Fostering Generalization in Single-View 3D Reconstruction by Learning a Hierarchy of Local and Global Shape Priors (CVPR 2021 · similarity 0.89)
- Deep Optimized Priors for 3D Shape Modeling and Reconstruction (CVPR 2021 · similarity 0.89)
- Learning View Priors for Single-View 3D Reconstruction (CVPR 2019 · similarity 0.88)
- Sparse Representation for 3D Shape Estimation: A Convex Relaxation Approach (TPAMI 2016 · similarity 0.87)
- Learning Shape Priors for Single-View 3D Completion and Reconstruction (ECCV 2018 · similarity 0.87)
- In Perfect Shape: Certifiably Optimal 3D Shape Reconstruction From 2D Landmarks (CVPR 2020 · 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.
