mesh-converter#

A static web app for converting meshes between formats: upload a mesh file, inspect it in an interactive 3D view (shaded, wireframe, shaded + wireframe, or points, with an optional anaglyph stereo mode for red/cyan 3D glasses), and download it in a format of your choice. Formats differ in what they can store, and the UI shows exactly which attributes (normals, colors) would be dropped by a lossy conversion.

Conversion is powered by meshio running in the browser via Pyodide — no server, all mesh I/O happens client-side. The first visit downloads the Python runtime (~15 MB from the jsDelivr CDN, cached afterwards).

† HDF5-based formats need the h5py Pyodide package (~4 MB); it is loaded lazily the first time such a format is used, so the default footprint stays small.

A loaded file is kept in its original bytes, in its original format — that is the source of truth. Export runs meshio directly on those bytes (read the original, write the target), so a conversion drops only what the target format genuinely cannot express, and exporting back to the same format returns the original file byte-for-byte with no conversion at all. During conversion the app translates vertex normals and colors between the formats' native representations (so PLY's nx/ny/nz properties become vn lines in OBJ, and so on), normalizes byte order and integer widths where meshio's writers are picky about dtypes, and strips data that meshio 5.3.5 would write corruptly (MDPA mesh data) or crash on (H5M and DOLFIN XML cell data). The 3D view is fed by a separate, triangle-only "common" representation: quads and polygons are fan-triangulated and volume cells are skipped for display, but none of that touches what gets exported.

✓ marks the attributes the app tracks for its loss warnings, and is how it writes the generated sample mesh (which, unlike an uploaded file, has no original bytes): vertex normals map to each format's native representation (nx/ny/nz scalars in PLY and Tecplot, vn lines in OBJ, a Normals point-data array elsewhere) and vertex colors likewise (red/green/blue in PLY and Tecplot, an RGB point-data array elsewhere).

Some meshio formats are deliberately absent: ansys, cgns, su2, and ugrid cannot roundtrip their own output in meshio 5.3.5; exodus fails writing under wasm; flac3d holds volume cells only; wkt's reader hangs on non-toy meshes; tetgen spans two files; svg is write-only.

Two built-in sample meshes are available from the UI for trying things out without a file: a generated rainbow torus with normals and vertex colors (examples/ holds it pre-exported in a few formats), and Keenan Crane's public-domain Spot cow, bundled as a real OBJ file so it exercises the upload path. An on-demand "estimate export sizes" action serializes the loaded mesh to every format in memory and shows the resulting file sizes in the format table and export dropdown.

A loaded mesh can be shared as a self-contained link: "Copy share link" packs the original uploaded file (byte-identical, in its original format) together with the chosen export format and view mode into the URL fragment — deflate-compressed and base64url-encoded after #share=. Nothing is uploaded; the data lives in the URL itself, so whoever opens the link sees the mesh and can download it in any format. Links are capped at 65,000 characters (roughly a 100–300 KB mesh file, depending on how well it compresses); beyond that the app says the mesh is too large to share by URL.

Development#

npm install
npm run dev      # local dev server
npm run build    # static build in dist/

Built with Vite, React, TypeScript, and three.js (react-three-fiber).

How it works#

• src/mesh/bridge.py runs inside Pyodide with three entry points: parse_mesh_file (native file → the triangle-only common form that feeds the viewer), convert_mesh (native → native, straight through meshio, the export path for uploaded files), and serialize_mesh (common form → native, used only for the generated sample). It exchanges arrays with JS as raw little-endian buffers through Pyodide's in-memory filesystem.
• src/mesh/meshio.ts loads Pyodide (script tag in index.html), installs meshio via micropip, and wraps the bridge in typed async functions: parseMeshFile for the viewer's MeshData (typed arrays of positions, triangle indices, optional normals/colors), convertMesh for lossless native-to-native export, and serializeMesh for the sample. App.tsx keeps the loaded file's original bytes as the source of truth and routes export through them — same-format exports skip meshio entirely.
• src/mesh/formats.ts declares the supported formats and which attributes each preserves; the upload, capability table, and loss-warning UI derive from it. To add a meshio-supported format, add a descriptor there (with pyodidePackages if it needs extra prebuilt packages such as h5py) and, if it stores normals/colors in a format-specific way, teach bridge.py how to map them.