This repository contains the implementation of the algorithm for reconstructing curves in non-Euclidean domains, as described in the paper Reconstructing Curves from Sparse Samples on Riemannian Manifolds.
Clone the project
git clone --recursive https://github.com/filthynobleman/curves-surf.git
Warning
For Windows users, Geometry Central could raise errors at compile time. To avoid it, change
deps/geometry-central/deps/CMakeLists.txt:160 by removing the option SYMBOLIC.
Please be sure that all the submodules are updated to the latest version
git submodule update --init --recursive --remote
Then build with CMake
mkdir build
cd build
cmake ..
cmake --build . --config release --parallel
The building process produces a single executable bin/CurvesSurf[.exe], which accepts a
single argument. The argument is a configuration file for a run, containing the following
attributes
| Attribute | Domain | Comments | Meaning |
|---|---|---|---|
mesh |
string | Mandatory | The path to a file containing a triangular mesh. |
samples |
string | Mandatory | The path to a file containing a newline separated list of indices of the mesh. |
output_prefix |
string | Optional, by default the basename of the input mesh. | The prefix for all the output files. |
use_sig |
boolean | Optional, by default true. | Wheter or not to use SIGDT connectivity in favor of the dual Voronoi only. |
multi_components |
boolean | Optional, by default true. | Whether or not to solve for a curve in every connected component of the graph. |
force_hamiltonian |
boolean | Optional, by default false. | Whether or not to edit the graph in order to force it in becoming Hamiltonian. |
distance |
{ "dijkstra", "heat", "exact" } |
Optional, by default "dijsktra". |
Which distance function is used by the algorithm (Dijkstra's algorithm, geodesics in heat, or exact geodesics). |
bias |
{ "none", "degree", "distance" } |
Optional, by default "none". |
Which edge choice must be prioritized in findining Hamiltonian cycles (random, smallest degree first, shortest edge first). |
An example configuration file can be found in examples/settings.json.
The algorithm procudes four output files:
<output_prefix>vertices.obj, containing the 3D coordinates of each sample;<output_prefix>voronoi.obj, containing the exact geodesic lines representing the dual Voronoi connectivity;<output_prefix>sigdt.obj, if optionuse_sigis given, the SIGDT connectivity is produced instead;<output_prefix>info.csv, containing some summary information like the curve length and the running time;<output_prefix>cycle.obj, containing the exact geodesic lines representing the reconstructed curve. If a Hamiltonian cycle is not found, this file is not produced.
The file visualization/pick_samples.py contains a Python script for Blender for helping in picking samples.
After selecting the desired mesh, enable the edit mode, shift+select the samples, and run the script.
This repository contains the data for reproducing the experiment depicted in the teaser figure of the paper.
The experiment can be reproduced by running the executable without arguments (from the build directory) or by giving the file examples/teaser.json as argument.
Alternatively, you can run the PowerShell script make_teaser.ps1 for carrying out the entire compilation process as well. The results from the teaser experiments will be produced at build/teaser-cycles.obj.
All the output files can be imported into Blender "as-is".
The vertices can be visualized using a Geometry Node modifier and plugging the Mesh to Points node between the input and the output.
The curves can be visualized with right click -> convert to -> curve and the giving them a depth from Object data properties -> geometry -> bevel -> depth.
This repository is distributed under the MIT license and makes use of the following dependencies:
- Eigen3, distributed under the MLP-2.0 license;
- Geometry Central, distributed under the MIT license;
- JSON for Modern C++, distributed under the MIT license.