Fast, clean mesh I/O for Python. Read and write 3D mesh files in one line - no hidden surprises, no silent data corruption.
pip install polyxiosimport polyxios as px
# Read any supported format
mesh = px.read("brain.vtk")
# Inspect
print(mesh.vertices.shape) # (n_verts, 3)
print(len(mesh.element_types)) # number of elements
# Write to a different format
px.write(mesh, "brain.ply")
px.write(mesh, "brain.vtp")Need binary output or format-specific options?
px.write(mesh, "brain.vtk", binary=True)
px.write(mesh, "brain.ply", binary=True, endian="little")polyxios comes with a command-line interface pxios to quickly fetch, list, convert, and visualize 3D models.
--verbose can be given on either side of the subcommand (e.g. pxios --verbose fetch bunny.obj or pxios fetch bunny.obj --verbose) to print debug logs and full tracebacks when a command fails.
pxios list: Lists all available remote or cached files, or registered formats. The three listing modes below are mutually exclusive.--local: Lists locally cached files (can filter by optional extension argument, e.g.pxios list obj --local).--extensions/--formats: Lists all formats and extensions available in the remote catalog.--codecs: Lists all formats supported by polyxios codecs.
pxios fetch <filename|extension>: Downloads and caches a single model file (e.g.,bunny.obj) or every model catalogued for an extension (e.g.,objor.obj).pxios convert <input_file> <output_file>: Converts a model file from one format to another directly in a single process.pxios viz <filename>: Visualizes a local or cached model file using the FURY library.--lines: Render line elements usingactor.lineinstead of rendering as a surface/point cloud.--points: Render strictly as a point cloud.
# List all fetchable remote models
pxios list
# Fetch a single model
pxios fetch bunny.obj
# Fetch every model catalogued for an extension
pxios fetch vtk
# Convert a mesh file
pxios convert bunny.obj bunny.vtk
# Visualize a model
pxios viz bunny.objFor large meshes (gigabytes of binary data), pass lazy=True. polyxios
memory-maps the file and only loads the pages you actually touch - the rest
stays on disk until needed.
# File is opened but data is not loaded into RAM yet
mesh = px.read("huge_brain.vtk", lazy=True)
# Only the vertices are pulled from disk here
first_vertex = mesh.vertices[0]
# Element connectivity is still on disk until you access itLazy loading is supported for binary .vtk, .ply, and .stl files. ASCII
formats load eagerly (the whole file must be parsed to extract values). Binary
STL lazy mode skips vertex deduplication - vertices are returned as-is (3 per
triangle), avoiding the extra pass over the data.
| Format | Extension | Read | Write | Lazy load |
|---|---|---|---|---|
| VTK Legacy | .vtk |
✓ | ✓ | binary only |
| VTK RectilinearGrid | .vtr |
✓ | ✓ | - |
| VTK PolyData | .vtp |
✓ | ✓ | - |
| Wavefront OBJ | .obj |
✓ | ✓ | - |
| Stanford PLY | .ply |
✓ | ✓ | binary only |
| STL | .stl |
✓ | ✓ | binary only |
| OFF | .off |
✓ | ✓ | ASCII + big-endian binary, ST/C/N variants → vertex/face attrs |
| Abaqus | .inp |
✓ | ✓ | - |
| AVS-UCD | .avs |
✓ | ✓ | - |
| Medit binary | .meshb |
✓ | ✓ | binary only |
| DOLFIN/FEniCS XML | .xml |
✓ | ✓ | - |
| FLAC3D | .f3grid |
✓ | ✓ | zones + faces, groups → element tags |
| Gmsh | .msh |
✓ | ✓ (v2) | ASCII v2 + v4.1, physical groups → element tags |
| Nastran | .bdf .nas .fem |
✓ | ✓ | free/small/large field read, free-field write with large-field GRID on request; .dat via fmt=".bdf" |
| Tecplot ASCII | .tec |
✓ | ✓ | FE zone, POINT + BLOCK packing, solution variables → vertex attrs; .dat via fmt=".tec" |
| SU2 | .su2 |
✓ | ✓ | ASCII, VTK element codes, boundary markers → element tags |
| TetGen | .ele+.node |
✓ | ✓ | paired files, 1-/0-based indices, boundary markers → vertex tags, region attrs |
| UGRID (AFLR) | .ugrid |
✓ | ✓ | ASCII, tri/quad surface + tet/pyramid/prism/hex volume, boundary tags → element tags |
| Netgen | .vol |
✓ | ✓ | ASCII, points/edges/faces/cells incl. quadratic, bcnr/matnr + names → element tags |
| Well-Known Text | .wkt |
✓ | ✓ | 2D padded to z=0, holes → element attrs, EWKT SRID dropped |
20 formats supported - more coming via the plugin system.
from polyxios.transforms import pipeline, merge, filter_element_type, remove_orphan_vertices
# Compose transforms into a single function
clean = pipeline(
filter_element_type(keep="triangle"),
remove_orphan_vertices,
)
result = clean(mesh)
# Merge two meshes into one
combined = merge(mesh_a, mesh_b)Any third-party package can teach polyxios to read and write a new format - no fork required, no pull request needed.
Step 1 - write a codec (two functions, nothing more):
# mypackage/abc_codec.py
from polyxios._registry import Codec
from polyxios._types import PolyData
def read(path, *, lazy=False) -> PolyData:
...
def write(poly: PolyData, path, **opts) -> None:
...
def register():
return ".abc", Codec(read, write)Step 2 - declare an entry point in your pyproject.toml:
[project.entry-points."polyxios.codecs"]
abc = "mypackage.abc_codec:register"After pip install mypackage, polyxios picks up .abc automatically -
no configuration, no restart needed:
mesh = px.read("model.abc") # works out of the boxClone the repo, then use spin to manage the development workflow:
pip install spin
spin setup # add upstream remote + install dev deps (libomp on macOS)
spin install # build Cython extensions and install
spin install -e # editable install (source changes reflected immediately)| Command | Description |
|---|---|
spin setup |
First-time setup: upstream remote, dev deps, OpenMP on macOS |
spin build |
Build with Meson/ninja |
spin install |
Regular install (compiled) |
spin install -e |
Editable install for development |
spin test |
Run the full test suite |
spin test -k <pattern> |
Run tests matching a name pattern |
spin lint |
ruff linter + formatter check + codespell |
spin lint --fix |
Auto-fix lint and formatting issues |
spin docs |
Build Sphinx documentation |
spin docs --clean |
Wipe _build/ before building |
spin docs --open |
Build and open docs in the browser |
spin clean |
Remove build artifacts and __pycache__ |
spin release <version> |
Cut a release: bump version, tag, push, start next dev cycle |
See docs/contributing.rst for commit message
conventions and the full contributor guide.
For the full release workflow see docs/development.rst.
- No silent data corruption - large mesh indices raise an error instead of truncating
- All element groups preserved - a face belonging to multiple tags stays in all of them
- Safe on untrusted files - header counts validated before any memory allocation
- Memory-efficient - lazy mmap loading for large binary files
- Works without a compiler - pure Python fallbacks included; Cython hot-paths optional
See LICENSE.