Store CAD features of the OctreeMeshing dataset

[sebmestrallet]

I thought the OctreeMeshing dataset from Gao et al. 2019 ¹ was only .obj triangulations, without CAD features such as CAD sharp edges.

Hovewer Beaufort et al. 2022 ² indicate (Table 1) that OctreeMeshing has 0D, 1D and 2D features for the subset of 109 CAD models. Section 2, pp 3-4:

The feature curves have been extracted using a dihedral angle of 140 degrees in combination with manual adjustment. A .fgraph file specifies the set of feature edges but without grouping them semantically into feature curves.

We should at least copy these .fgraph files while importing OctreeMeshing in definitions/algorithms/import_OctreeMeshing.py, or even better read these files (what is the structure of a .fgraph file?) and merge them with corresponding triangulation, to have both the mesh and the CAD features in a same file (.geogram?), for each model.

Not taking into account CAD features degraded the results of the algorithm described in our last article ³, and removed one metric to compare it to state of the art.

Footnotes

1. X. Gao, H. Shen, and D. Panozzo, “Feature Preserving Octree-Based Hexahedral Meshing”, Computer
   Graphics Forum, vol. 38, no. 5, pp. 135–149, 2019, doi:10.1111/cgf.13795 ↩

2. P.-A. Beaufort, M. Reberol, D. Kalmykov, H. Liu, F. Ledoux, and D. Bommes, “Hex me if you can”, Computer Graphics Forum, vol. 41, no. 5, pp. 125–134, 2022, doi:10.1111/cgf.14608 ↩

3. S. Mestrallet, C. Bourcier, and F. Ledoux, "Validity-first automatic polycube labeling for CAD models", Proceedings of the 2025 International Meshing Roundtable, SIAM, March 2025, arXiv:2502.05712, GitHub:LIHPC-Computational-Geometry/validity-first-polycube-labeling ↩