Research map / 3D vision / Face and shape
Neural Fields As Learnable Kernels for 3D Reconstruction
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 a learned, input-conditioned kernel with closed-form ridge regression rather than the compared reconstruction methods | ShapeNet reconstruction · ShapeNet noise-free | IoU: 0.864 -> 0.949 | Helped |
| Use a learned, input-conditioned kernel with closed-form ridge regression rather than the compared reconstruction methods | NKF · ShapeNet noise-free | Chamfer: 0.048 -> 0.024 | Helped |
| Use NKF rather than C-OccNet for partial-shape completion | ShapeNet completion · ShapeNet completion | IoU: 0.770 -> 0.819 | Helped |
| Use NKF rather than C-OccNet for partial-shape completion | ShapeNet completion · ShapeNet completion | Chamfer: 0.075 -> 0.077 | Mixed |
| Train on 6 categories and evaluate on unseen categories using NKF | NKF · ShapeNet out-of-category | IoU: 0.869 -> 0.938 | Helped |
| Apply NKF to scanned scenes instead of the listed baselines | ScanNet reconstruction · ScanNet | Chamfer: 0.060 -> 0.032 | Helped |
| Add the surface L1 loss (ablation) | NKF, feature dimension 32 · ShapeNet test set | IoU: 0.942 -> 0.949 | Helped |
| Add the surface L1 loss (ablation) | NKF, feature dimension 8 · ShapeNet test set | IoU: 0.939 -> 0.945 | Helped |
| Use feature dimension 8 rather than 32, with surface L1 loss (ablation) | NKF · ShapeNet test set | IoU: 0.949 -> 0.945 | Hurt |
| Use feature dimension 16 rather than 32, with surface L1 loss (ablation) | NKF · ShapeNet test set | IoU: 0.949 -> 0.947 | Hurt |
| Use feature dimension 64 rather than 32, with surface L1 loss (ablation) | NKF · ShapeNet test set | IoU: 0.949 -> 0.949 | No clear effect |
| Use feature dimension 16 rather than 8, without surface L1 loss (ablation) | NKF · ShapeNet test set | IoU: 0.939 -> 0.941 | Helped |
Closest papers by meaning
- Neural Kernel Surface Reconstruction (CVPR 2023 · similarity 0.93)
- Neural Vector Fields: Implicit Representation by Explicit Learning (CVPR 2023 · similarity 0.89)
- Robustifying Generalizable Implicit Shape Networks with a Tunable Non-Parametric Model (NeurIPS 2023 · similarity 0.88)
- Neural Parts: Learning Expressive 3D Shape Abstractions With Invertible Neural Networks (CVPR 2021 · similarity 0.88)
- Neural Splines: Fitting 3D Surfaces With Infinitely-Wide Neural Networks (CVPR 2021 · similarity 0.88)
- Learning Partonomic 3D Reconstruction from Image Collections (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.