Research map / 3D vision / Face and shape
OperatorNet: Recovering 3D Shapes From Difference Operators
Ruqi Huang, Marie-Julie Rakotosaona, Panos Achlioptas, Leonidas J. Guibas, Maks Ovsjanikov
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 extrinsic shape difference to intrinsic shape differences (ablation) | OperatorNet · DFAUST test | d_R (x1e-4): 2.41 -> 1.11 | Improved |
| add intrinsic shape differences to extrinsic shape difference (ablation) | OperatorNet · DFAUST test | d_R (x1e-4): 1.25 -> 1.11 | Improved |
| use ground-truth correspondences instead of computed functional maps (Int+Ext) (ablation) | OperatorNet · DFAUST test | d_R (x1e-4): 6.22 -> 1.11 | Improved |
| use ground-truth correspondences instead of computed functional maps (Ext) (ablation) | OperatorNet · DFAUST test | d_R (x1e-4): 3.86 -> 1.25 | Improved |
| replace SfL baseline with OperatorNet (Int+Ext) | OperatorNet · DFAUST test | d_R (x1e-4): 48.8 -> 1.11 | Improved |
| replace nearest neighbor retrieval with OperatorNet (Int+Ext) | OperatorNet · DFAUST test | d_V: 0.005 -> 0.014 | Hurt |
| replace SfL (full Laplacian input) with OperatorNet (Int+Ext) | OperatorNet · DFAUST test | d_E: 0.012 -> 0.045 | Hurt |
| add area and conformal differences to extrinsic (all three vs Area only) (ablation) | OperatorNet, Conv. 8 · DFAUST test | reconstruction loss (x1e-4): 8.61 -> 2.46 | Improved |
| add area difference to E+C (ablation) | OperatorNet, Conv. 8 · DFAUST test | reconstruction loss (x1e-4): 2.56 -> 2.46 | Improved |
| use a shallower single-layer conv encoder (Conv. 8) instead of four layers (ablation) | OperatorNet, A+E+C · DFAUST test | reconstruction loss (x1e-4): 4.24 -> 2.46 | Improved |
| use a single-layer encoder instead of two layers (ablation) | OperatorNet, A+E+C · DFAUST test | reconstruction loss (x1e-4): 3.05 -> 2.46 | Improved |
| use multiplicative (Lie algebra) interpolation instead of linear interpolation of shape differences (ablation) | OperatorNet · shape interpolation | smoothness of consecutive reconstructions | Improved |
Closest papers by meaning
- Higher-Order Function Networks for Learning Composable 3D Object Representations (ICLR 2020 · similarity 0.89)
- Deep Meta Functionals for Shape Representation (ICCV 2019 · similarity 0.88)
- Neural Parts: Learning Expressive 3D Shape Abstractions With Invertible Neural Networks (CVPR 2021 · similarity 0.88)
- Deep Geometric Functional Maps: Robust Feature Learning for Shape Correspondence (CVPR 2020 · similarity 0.87)
- Learning Partonomic 3D Reconstruction from Image Collections (CVPR 2025 · similarity 0.86)
- Denoising Functional Maps: Diffusion Models for Shape Correspondence (CVPR 2025 · similarity 0.86)
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.
