Research map / 3D vision / Face and shape
FoV-Net: Rotation-Invariant CAD B-rep Learning via Field-of-View Ray Casting
Matteo Ballegeer, Dries F. Benoit
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 |
|---|---|---|---|
| use rotation-invariant FoV-Net instead of UV-Net under random rotation | UV-Net · SolidLetters (rotated test) | accuracy (%): 8.94 ± 0.46 -> 96.35 ± 0.16 | Improved |
| use FoV-Net instead of AAGNet under random rotation | AAGNet · MFCAD++ (rotated test) | per-face accuracy (%): 80.13 ± 1.71 -> 99.33 ± 0.07 | Improved |
| use FoV-Net instead of UV-Net under random rotation | UV-Net · Fusion360 (rotated test) | per-face accuracy (%): 69.13 ± 2.18 -> 91.72 ± 0.17 | Improved |
| add rotation augmentation to UV-Net | UV-Net · SolidLetters (rotated test) | accuracy (%): 8.94 ± 0.46 -> 94.97 ± 0.10; original 97.10 ± 0.15 -> 95.34 ± 0.12 | Mixed |
| add rotation augmentation to AAGNet | AAGNet · Fusion360 | per-face accuracy (%): rotated 79.85 ± 0.92 -> 86.70 ± 0.28; original 90.60 ± 0.25 -> 87.91 ± 0.56 | Mixed |
| add global UV grids to FoV-Net (replace LRF UV) (ablation) | FoV-Net · SolidLetters | accuracy (%): original 96.35 ± 0.16 -> 97.42 ± 0.17; rotated 96.35 ± 0.16 -> 23.53 ± 0.61 | Mixed |
| add global UV grids instead of LRF UV to FoV-Net (ablation) | FoV-Net · MFCAD++ (rotated test) | per-face accuracy (%): 99.33 ± 0.07 -> 70.45 ± 3.20 | Hurt |
| combine LRF UV with FoV grids (ablation) | FoV-Net · SolidLetters | accuracy (%): 94.39 -> 96.35 | Improved |
| combine OV and IV grids (ablation) | FoV-Net · SolidLetters | accuracy (%): 92.92 -> 95.79 | Improved |
| add FoV and LRF descriptors over per-face features alone (ablation) | FoV-Net · SolidLetters | accuracy (%): 70.68 -> 96.35 | Improved |
| add descriptors over topology alone (ablation) | FoV-Net · SolidLetters | accuracy (%): 37.72 -> 96.35 | Improved |
| increase FoV grid from 1x1 single ray to multi-directional rays (ablation) | FoV-Net (FoV only) · SolidLetters | accuracy (%): 75.04 -> 95.79 | Improved |
Closest papers by meaning
- SO(3)-Equivariant Learning on CAD Boundary Representations (NeurIPS 2026 · similarity 0.90)
- UV-Net: Learning From Boundary Representations (CVPR 2021 · similarity 0.90)
- Self-Supervised Representation Learning for CAD (CVPR 2023 · similarity 0.89)
- BRepFacetGen: Reverse Engineering B-Reps By Generative Face Segmentation (ECCV 2026 · similarity 0.87)
- SpelsNet: Surface Primitive Elements Segmentation by B-Rep Graph Structure Supervision (NeurIPS 2024 · similarity 0.86)
- PolyhedronNet: Representation Learning for Polyhedra with Surface-attributed Graph (ICLR 2025 · 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.
