Research map / 3D vision / Face and shape
Neural Vector Fields: Generalizing Distance Vector Fields by Codebooks and Zero-Curl Regularization
Xianghui Yang, Guosheng Lin, Zhenghao Chen, Luping Zhou
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 hard codebook (ablation) | NVF (ultra) · ShapeNet non-watertight (base classes) | Normal error: 0.196 -> 0.162 | Improved |
| add hard codebook (ablation) | NVF (ultra) · ShapeNet non-watertight (novel classes) | Normal error: 0.193 -> 0.169 | Improved |
| replace hard codebook with soft codebook (ablation) | NVF (ultra) · ShapeNet non-watertight (base classes) | Normal error: 0.162 -> 0.143 | Improved |
| replace hard codebook with soft codebook (ablation) | NVF (ultra) · ShapeNet non-watertight (novel classes) | Normal error: 0.169 -> 0.153 | Improved |
| add curl regularization to soft codebook (ablation) | NVF (ultra) · ShapeNet non-watertight (base classes) | Normal error: 0.143 -> 0.144 | No clear effect |
| add curl regularization to soft codebook (ablation) | NVF (ultra) · ShapeNet non-watertight (novel classes) | Chamfer distance: 0.818 -> 0.841 | Hurt |
| add direction loss to soft codebook plus curl (ablation) | NVF (ultra) · ShapeNet non-watertight (base classes) | Normal error: 0.144 -> 0.137 | Improved |
| add direction loss to soft codebook plus curl (ablation) | NVF (ultra) · ShapeNet non-watertight (novel classes) | Chamfer distance: 0.841 -> 0.745 | Improved |
| use PointTransformer backbone instead of 3D convolution (ablation) | NVF (ultra) · ShapeNet non-watertight (base classes) | Chamfer distance: 0.886 -> 0.810 | Improved |
| use PointTransformer backbone instead of 3D convolution (ablation) | Ours (lite) · ShapeNet non-watertight (base classes) | Chamfer distance: 0.92 -> 0.85 | Improved |
| use soft codebook with curl and direction loss (ultra) instead of hard codebook (lite) | NVF · MGN (cross-domain) | Chamfer distance: 0.141 -> 0.133 | Improved |
| use NVF (lite) instead of NDF for inference | NVF · 50k-query inference | runtime: 0.327s -> 0.113s | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- GIFS — GIFS: Neural Implicit Function for General Shape Representation (CVPR 2022) · better on 7 of 7 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- 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)
- A Lightweight UDF Learning Framework for 3D Reconstruction Based on Local Shape Functions (CVPR 2025 · similarity 0.88)
- Small Steps and Level Sets: Fitting Neural Surface Models with Point Guidance (CVPR 2024 · similarity 0.88)
- DUDF: Differentiable Unsigned Distance Fields with Hyperbolic Scaling (CVPR 2024 · similarity 0.88)
- Details Enhancement in Unsigned Distance Field Learning for High-fidelity 3D Surface Reconstruction (AAAI 2025 · 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.
