Research map / Point clouds / Shape and surface completion
NeuroGF: A Neural Representation for Fast Geodesic Distance and Path Queries
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 |
|---|---|---|---|
| Use NeuroGF instead of the heat method for SSAD on anisotropic dragon and bucket meshes | SSAD geodesic queries · dragon and bucket | mean relative error (%): 10.6 -> 0.68; 3.35 -> 0.18 | Helped |
| Use NeuroGF rather than fDGG for SSAD query time | SSAD geodesic queries · tested meshes | query time (ms): fDGG 4–145; NeuroGF 0.4–0.7 | Helped |
| Add signed-distance-field supervision branch (ablation) | NeuroGF distance representation · all testing shapes | mean relative error (%): 0.57 -> 0.49 | Helped |
| Add signed-distance-field supervision branch (ablation) | NeuroGF path representation · all testing shapes | Chamfer-L1 (×10^-2): 1.31 -> 1.25 | Helped |
| Add curve-length consistency loss (ablation) | NeuroGF distance representation · all testing shapes | mean relative error (%): 0.52 -> 0.49 | Helped |
| Add curve-length consistency loss (ablation) | NeuroGF path representation · all testing shapes | Chamfer-L1 (×10^-2): 1.35 -> 1.25 | Helped |
| Add curve-point distribution constraint (ablation) | NeuroGF path representation · all testing shapes | Chamfer-L1 (×10^-2): 1.38 -> 1.25 | Helped |
| Increase the number of shape encoder cues by using GraphConv rather than PointTr | Generalizable NeuroGF · SN-8x50 and SN-5x50 | MRE (%): 3.28 -> 2.94; 4.16 -> 3.55 | Helped |
| Increase ground-truth training source count on dragon (ablation) | NeuroGF offline training · dragon | MRE (%): 64: 1.63; 1024: 0.68; 10000: 0.66 | Helped |
| Train with only 8K rather than original ground-truth pairs (ablation) | NeuroGF training data · unspecified testing set | MRE (%): 0.53 -> 4.05 | Hurt |
| Train with only 2K rather than original ground-truth pairs (ablation) | NeuroGF training data · unspecified testing set | MRE (%): 0.53 -> 5.62 | Hurt |
Closest papers by meaning
- De-coupled NeuroGF for Shortest Path Distance Approximations on Large Terrain Graphs (ICML 2025 · similarity 0.90)
- Implicit-ARAP: Efficient Handle-Guided Neural Field Deformation via Local Patch Meshing (NeurIPS 2025 · similarity 0.86)
- Neural Processing of Tri-Plane Hybrid Neural Fields (ICLR 2024 · similarity 0.85)
- Grounding Continuous Representations in Geometry: Equivariant Neural Fields (ICLR 2025 · similarity 0.85)
- DUDF: Differentiable Unsigned Distance Fields with Hyperbolic Scaling (CVPR 2024 · similarity 0.85)
- Flatten Anything: Unsupervised Neural Surface Parameterization (NeurIPS 2024 · similarity 0.85)
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.