Research map / 3D vision / Face and shape
Neural Vector Fields: Implicit Representation by Explicit Learning
Read the paperCheck 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 |
|---|---|---|---|
| Represent query locations with learned displacement vector fields rather than scalar unsigned distance fields | 3D surface reconstruction · ShapeNet Cars, category-specific | CD: 0.146 -> 0.114 | Helped |
| Represent query locations with learned displacement vector fields rather than scalar unsigned distance fields | 3D surface reconstruction · ShapeNet non-watertight, novel classes | CD: 0.093 -> 0.078 | Helped |
| Use vector fields for cross-domain reconstruction | ShapeNet-trained models evaluated on MGN · MGN cross-domain | CD: 0.025 -> 0.014 | Helped |
| Add the multi-head codebook (ablation) | NVF with K=16 · ShapeNet non-watertight, base classes | CD: 0.121 -> 0.085 | Helped |
| Add the multi-head codebook (ablation) | NVF with K=16 · ShapeNet non-watertight, novel classes | CD: 0.081 -> 0.078 | Helped |
| Increase the number of nearest-point features from K=8 to K=16 with the codebook (ablation) | NVF · ShapeNet non-watertight, base classes | CD: 0.087 -> 0.085 | Hurt |
| Increase the number of nearest-point features from K=8 to K=16 with the codebook (ablation) | NVF · ShapeNet non-watertight, novel classes | CD: 0.083 -> 0.078 | Hurt |
| Replace PointTransformer with a point-cloud backbone rather than 3D Conv (ablation) | NVF · ShapeNet non-watertight, base and novel classes | CD: 0.092 -> 0.085; 0.090 -> 0.078 | Mixed |
| Use PointTransformer instead of PointNet++ (ablation) | NVF · ShapeNet non-watertight, base classes | EMD: 1.198 -> 1.197 | No clear effect |
| Use NVF instead of NDF with the same 3D Conv backbone (ablation) | 3D surface reconstruction inference · 200k queries | Runtime: 0.75s -> 0.27s | Helped |
| Add the multi-head codebook to NVF with 3D Conv (ablation) | NVF inference · 200k queries | Runtime: 0.27s -> 0.34s | Hurt |
| Add the multi-head codebook to NVF with PointTransformer (ablation) | NVF inference · 200k queries | Runtime: 0.29s -> 0.35s | Hurt |
Closest papers by meaning
- Neural Unsigned Distance Fields for Implicit Function Learning (NeurIPS 2020 · similarity 0.90)
- Unsigned Orthogonal Distance Fields: An Accurate Neural Implicit Representation for Diverse 3D Shapes (CVPR 2024 · similarity 0.89)
- 3D Neural Edge Reconstruction (CVPR 2024 · similarity 0.89)
- An Intrinsic Vector Heat Network (ICML 2024 · similarity 0.88)
- Learning 3D Equivariant Implicit Function with Patch-Level Pose-Invariant Representation (NeurIPS 2024 · similarity 0.88)
- DUDF: Differentiable Unsigned Distance Fields with Hyperbolic Scaling (CVPR 2024 · 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.