This repository contains the implementation of Atomizer, a novel toolpath generation algorithm for fused filament fabrication (FFF) 3D printing. Atomizer introduces the concept of atoms as a replacement for traditional slices. This method enables collision-free toolpaths that tightly conform to complex geometries and offer new fabrication capabilities.
Xavier Chermain, Giovanni Cocco, Cédric Zanni, Eric Garner, Pierre-Alexandre Hugron, and Sylvain Lefebvre
Université de Lorraine, CNRS, Inria, Loria
Computer Graphics Forum (Proceedings of the Symposium on Geometry Processing), 2025
SGP Honorable Mention
- Toolpath generation using frames (i.e., atoms) instead of slices
- Control of the deposition direction and tool orientation
- Collision-free toolpath ordering
- Support for anisotropic appearance on curved surfaces
git clone https://github.com/xavierchermain/atomizer
This implementation is written in Python 3.10. Supported versions are Python >= 3.7 and < 3.11, as required by the Taichi module dependency. The implementation also relies on the local library src/atom. Please refer to pyproject.toml for the complete list of Python module dependencies.
We recommend using Miniconda to install a specific version of Python, the dependencies, and the local library src/atom:
conda create --name atomizer python=3.10
conda activate atomizer
conda install pip
pip install --user -e .
Blender is also a dependency. You need to install it and add it to your executable path. It is used to remesh the model (see tools/process_for_atomizer.py) for the atomizer. We use Blender 4.4.
python tools/atomize.py data/param/<json_filename>.json
The parameters in the JSON file are:
-
solid_name: the filename of the STL representing the 3D solid, without the extension. The STL file must be located in thedata/mesh/folder and have the.stlextension. -
deposition_width: the deposition width in millimeters. The layer height is half the deposition width in our implementation. -
max_slope: the maximum tilting angle of the tool, in degrees. -
top_linesandbottom_lines(optional): paths to single-channel, 8-bit-per-pixel PNG files representing the target tangent directions for the top and bottom surfaces, respectively. The mapping is$[0, 255] \leftarrow [−\pi/2, \pi/2]$ . The orientation defines a 2D line in the xy-plane. The 2D line field is defined on the upper face of the solid’s bounding box and is planarly projected onto the top and bottom surfaces along the z-axis. -
ortho_to_wall(optional): if true, forces the tool orientation to be parallel to the boundary. By default, this is false, as enabling this feature causes many tool orientation changes that are detrimental to surface quality.
The inputs and outputs are:
- STL input:
data/mesh/<json_filename>.stl - Toolpath output:
data/toolpath/<json_filename>.npz - Log file:
data/log/<json_filename>.log
Refer to the log file to see all the individual computation and visualization commands.
Add the --warmup option to exclude the compilation time from the computation time reported in the log file. Caution: this causes each step of the pipeline to run twice, as Taichi uses just-in-time compilation.
python tools/atomize.py data/param/triangle_24.json
After using the atomizer, you can visualize the generated <toolpath> with:
python tools/visualize_toolpath.py data/toolpath/<toolpath>.npz
Use the WASD keys to move the camera. Press H to hide the travel paths.
Check the log file generated by the atomizer tool to see all individual visualization commands.
python tools/visualize_toolpath.py data/toolpath/triangle_24_smoothed.npz
python tools/toolpath_to_ply.py data/toolpath/<toolpath>.npz data/mesh/<toolpath>.ply
The generated mesh <toolpath>.ply can be visualized in Blender. To display the toolpath relative length using the Turbo colormap, you need to add a material to the mesh. Then, in the Material tab, do the following:
- Select Base Color.
- Choose Color Attribute.
To visualize the vertex color attribute:
- Select Shading (shortcut z).
- Choose Material Preview (shortcut 2).
Tool manuals are accessible by typing --help, e.g.,
python tools/atomize.py --help
atomizer/
├── tools/ # The main tools
│ │── atomize.py # Main pipeline entry point
│ │── bpn_to_sdf.py # SDF generation and voxelization
│ │── compute_tool_orientations.py # Tool orientation field optimization
│ │── compute_tangents.py # Deposition tangent field optimization
│ │── align_atoms.py # Tricosine field optimization
│ │── extract_explicit_atoms.py # Atom extraction
│ │── order_atoms.py # Atom ordering
│ │── visualize_bpn_sdf.py # Visualize SDF
│ │── visualize_bpn_df.py # Visualize tool orientation field
│ │── visualize_bases.py # Visualize deposition tangent field
│ │── visualize_implicit_atoms.py # Visualize tricosine field
│ │── visualize_explicit_atoms.py # Visualize explicit atoms
│ │── visualize_toolpath.py # Visualize generated toolpath
│ └── ... # Other tools
│── data/ # Input and output data
│ │── mesh/ # Input STLs
│ │── param/ # Input parameter JSON files
│ │── toolpath/ # Output toolpaths
│ └── ... # Other data
├── src/atom/ # Local library with core functionalities
├── experiment/ # For experimenting with core functionalities
├── generate_results.ps1 # Script to generate toolpaths
└── README.md
Atomizer was tested on a custom 3-axis printer with independently controlled Z-axis screws. The customization is intended for experts; consequently, we do not provide any GCODE to prevent users from damaging their machines. The code generates only the sequence of positions, each associated with a tool orientation and a travel type (deposition or no deposition). The generated toolpaths are available in data/toolpath. To print, you must use the inverse kinematics model of your machine.
If you use this code in your research, please cite:
@article{Chermain2025Atomizer,
author = {Chermain, Xavier and Cocco, Giovanni and Zanni, Cédric and Garner, Eric and Hugron, Pierre-Alexandre and Lefebvre, Sylvain},
title = {{Atomizer: Beyond Non-Planar Slicing for Fused Filament Fabrication}},
journal = {Computer Graphics Forum (Proceedings of the Symposium on Geometry Processing)},
year = {2025}
doi = {https://doi.org/10.1111/cgf.70189},
}
The source code is under the BSD 3-Clause "New" or "Rivised" license. See LICENSE for more details.