Primtag

Setting Up the Python Environment

First, download OpenSCAD from here.

For Mac or Linux users, add the following line to your .zshrc or .bashrc, depending on the shell you're using:

export OPENSCAD_EXEC=<full_path_to_openscad_executable>

For Windows users, add an environment variable named OPENSCAD_EXEC with the path to the OpenSCAD executable.

Ensure the modules variable is updated to the correct absolute path for your system.

Then, install the necessary packages in your Python environment:

pip install solidpython
pip install libigl

A working Python version is 3.8, but the latest version should also be compatible.

Using fabhacks_min

Activate your environment and run:

python viewer

Accessing the libigl Repository

Clone the libigl repository to your local machine:

git clone https://github.com/libigl/libigl.git

Build libigl with the following commands:

cd libigl
mkdir build
cd build
cmake ..
make

To run the tutorial files, for example, 106_ViewerMenu.cpp, navigate to the bin directory and run:

./106_ViewerMenu

For more detailed instructions on using the libigl repository, visit their tutorial page.

Building UI for Mesh Segmentation

In the step2 directory, 106_ViewerMenu.cpp has been enhanced with additional functionalities, such as tracking mouse coordinates, plotting screen points, and segmenting the mesh for user interaction with the uploaded mesh. This directory also contains the original and reduced meshes.

We use the libigl repository to segment a primitive mesh (reduced mesh) from the original mesh and save it as an OBJ file. Follow these steps:

Upload a mesh.
Click the track mouse button to start tracking the mouse's screen coordinates.
Use the plot points button to plot intersection points and the color faces button to visualize the hit faces.
Click the segmentation button to segment and save the mesh from the original mesh based on the selected region.

Ensure that the selected region is precise and forms a tight boundary for the desired primitive type.

Parameters Optimization

In the step3_parameters_learning directory, we include seven different types of primitives along with corresponding parameter learning .ipynb files.

The Python environment for these Jupyter Notebook files requires Python 3.10, along with PyTorch and PyTorch3D dependencies.

Limitations

Due to the instability of randomly sampling points on the surface of each reduced mesh, the learning part for Clip, Hook, and Hole might not be accurate at all times. If the results are incorrect, rerun the file.

There is a slight discrepancy between our current techniques and the ground truth for some meshes with small sizes. However, as the size of the mesh increases, our techniques become more accurate and have a smaller margin of error.

Future Work

We have uploaded the volumetric mesh, which includes interior faces, for the reduced mesh in the step2 directory. In future work, we aim to reconstruct the parametrization function for each primitive to sample points both on the surface and interior, improving the accuracy and stability of the learning part.

Research Showcase

Our team has presented our research poster in two research showcase sessions. The poster highlights our findings and the methodologies used in the project. You can view the poster below:

merlinyx/primtag