Voronizer — Open-Cell Strut Generator

Voronizer turns any STL mesh into an airy, open-cell structure built from Voronoi struts. CUDA kernels voxelize the source model, seed a Voronoi diagram, and emit printable lattices for both the model interior and optional support scaffolding. The latest iteration focuses on hollow, strut-based Voronization—dial in shell thickness, strut diameters, and surface nets to produce lightweight foams straight from your GPU (or fall back to a CPU path when CUDA is unavailable).

What You Get

• Open Voronoi foam: fill the model volume with struts or keep only a thin surface net.
• Per-feature control: independent parameters for model infill and support lattices (MODEL_CELL, SUPPORT_CELL, thresholds, shells).
• GPU-accelerated pipeline: numba.cuda kernels for voxelization, signed-distance evaluation, and Voronoi carving.
• Debug + export tools: slice visualizer, raw voxel analysis, and mesh exporters for downstream cleanup.

Requirements

• NVIDIA GPU + CUDA Toolkit (10.x or newer recommended). When CUDA is missing, Voronizer automatically falls back to a CPU implementation of the strut finder—expect noticeably longer runtimes on large grids (though tiny lattices can actually finish faster on CPU because they avoid CUDA launch/transfer overhead).
• Python 3.8+ with numba, numpy, matplotlib, Pillow, scikit-image installed.

pip install numba numpy matplotlib Pillow scikit-image

Repository Layout

Path	Purpose
Input/	Demo STLs; add your own meshes here.
Output/	Runtime export bucket (ignored by git). Meshes arrive as .ply.
userInput.py	All toggles: select STL, resolution, strut diameters, shell thickness, perforations, etc.
main.py	Pipeline driver: voxelize → Voronize → smooth → export.
voxelize.py, voronize.py, Frep.py, SDF3D.py	CUDA kernels and helpers for SDF + Voronoi math.
visualizeSlice.py, analysis.py	Optional helpers for debugging cross-sections and metrics.

Quick Start

1. Drop an STL in Input/.
2. In userInput.py set FILE_NAME = "myMesh.stl" and tweak:
   • RESOLUTION: grid density (start ~140 for testing).
   • MODEL_CELL / SUPPORT_CELL: target Voronoi strut thickness in voxels.
   • MODEL_THRESH / SUPPORT_THRESH: cell density. Larger = more points/struts.
   • MODEL_SHELL: add a solid skin outside the lattice (keep 0 for fully open cells).
   • NET = True + NET_THICKNESS: surface-only webbing.
   • PERFORATE = True: drill holes through support cells for resin flushes.
3. Run python main.py. On systems without CUDA, the CPU fallback kicks in automatically; just budget extra time for the strut-finding phase (unless you’re experimenting on very small structures, where the CPU route can be quicker).
4. Collect .ply meshes from Output/ and, if necessary, post-process in MeshLab (clean non-manifold faces, apply HC-Laplacian smoothing).

Open-Cell Tips

• Balance MODEL_CELL with RESOLUTION: smaller voxels allow thinner struts without breakup.
• For delicate lattices, keep SMOOTH = True and consider AESTHETIC = True to remove internal clutter while keeping the outer foam.
• Need two printable parts? Set SEPARATE_SUPPORTS = True to emit model + supports independently.
• Keep GPU memory headroom: if you see CudaAPIError: cuMemcpyDtoH, drop RESOLUTION or simplify the STL (MeshMixer → Simplification → 0.5 reduction works well).

Troubleshooting

• UnicodeDecodeError while loading STL → Re-export your mesh as a binary STL.
• Runtime exceeds minutes → Lower RESOLUTION, decimate the mesh, and close other GPU-hungry apps.
• Banding or chunky struts → Increase SMOOTH_STEPS in main.py or run HC-Laplacian smoothing in MeshLab after export.

Contributing

• Keep GPU kernels colocated with their host helpers (e.g., CUDA bits stay in voxelize.py, Frep.py).
• Add lightweight assertions/logging for new utilities so regressions surface quickly.
• Output directories (Output/, caches, .ply) are ignored; please keep the repo clean of generated meshes.

Create, tweak, and print latticed art—happy Voronizing!