Convert glTF to PLY

PLY came out of Stanford in 1994 for exactly this purpose: a mesh with an arbitrary, self-describing set of per-vertex and per-face properties, in a header a program can read without knowing anything in advance. That design is why it became the default currency of scanning, photogrammetry and academic geometry work, and why CloudCompare, MeshLab, Open3D and trimesh all treat it as a first-class input.

glTF was designed for the opposite end of the pipeline — shipping a finished asset to a renderer over a network. Its strengths are the things a measurement tool ignores: material definitions, texture packing, animation, extension mechanisms. Several of those tools have no glTF importer at all, and the ones that do treat it as an afterthought. Converting is usually faster than fighting that.

It is deliberately minimal: the magic word ply, then format ascii 1.0, a comment naming the writer, an element vertex declaration with its count, three property float lines for x, y and z, an element face declaration with its count, one property list uchar int vertex_indices line, and end_header.

That is the most widely supported PLY dialect there is. A uchar face count with int indices is what nearly every reader assumes when it takes shortcuts, and float rather than double positions keeps the file half the size for a precision no mesh here needs. If you are writing a parser against this output, those seven declarations are the entire contract.

Every coordinate is written as decimal text, one vertex per line and one triangle per line. A vertex that occupied twelve bytes in the glTF buffer occupies twenty to sixty characters here, so expect the PLY to be several times the size of the geometry it came from. A two-million-triangle model is a large text file.

The compensation is that nothing about it is ambiguous. Binary PLY comes in little-endian and big-endian variants with property layouts a reader has to reconstruct from the header, and mismatched assumptions there produce meshes that load as noise. ASCII cannot fail that way, and for a file that is going into an analysis tool once, the size rarely matters more than the certainty.

A PLY file holds one vertex element and one face element. There is no group, no object and no part. The converter therefore merges every mesh it read from the glTF scene into a single list, offsetting each mesh's indices by the number of vertices already written so the triangles still point at the right places.

The geometry is unchanged and the structure is gone. A glTF assembled from twelve named parts becomes one surface of twelve disconnected shells, and the names go with the structure. If those divisions are what you need downstream, OBJ is the target that keeps them, and converting to OBJ instead costs you nothing else that PLY would have provided.

The single most cited reason to use PLY is that it holds red, green and blue per vertex, which is how a photogrammetry scan carries its appearance without a texture. The format supports it and readers expect it.

This converter writes none. Colour in a glTF lives in materials and textures indexed by UV coordinates, and none of those is read; there is no vertex colour attribute to carry through unless the source declared one, and that attribute is not read either. The output is a colourless surface, which is correct for measurement and wrong if what you wanted was an appearance-preserving export.

The scene is walked from its default scene, each node matrix multiplied by its parent, and the world matrix applied to the vertices. Nodes authored with separate translation, rotation and scale are composed into a matrix first, following the quaternion convention glTF defines.

Doing this before the merge is what makes the merge safe. If transforms were left unapplied, flattening several meshes into one list would stack every part at the origin and produce a shape that has no relationship to the model. Baking first means the numbers you measure in CloudCompare are the numbers the glTF scene described.

An index buffer, carried through verbatim. glTF stores triangles as indices into a shared position array, and the conversion preserves that structure element for element rather than reconstructing it — no merging is applied on this path, because there is nothing to repair. An STL arriving at the same destination has to be welded first, since it stores three unshared vertices for every triangle by design.

That distinction decides whether the analysis you are about to run means anything. Manifold checks, normal estimation, curvature, geodesic distance, Laplacian smoothing and decimation all assume connectivity. The one caveat worth carrying into MeshLab is that exporters split vertices at UV and normal seams, and those duplicates survive here as separate entries at identical coordinates, so a Merge Close Vertices pass before a topological check is a sensible habit.

A .gltf that keeps its geometry in a separate .bin cannot be read, because a single-file converter never sees the second file; the message says so and suggests the .glb, which has everything in one place. A file using KHR_draco_mesh_compression stops with a message naming Draco, whose decoder is not shipped here.

Primitives that are not triangle lists are skipped. glTF mode 4 is triangles; strips, fans, lines and points are not surfaces this converter can turn into faces, and a file containing only those reports that no triangle geometry was found rather than producing an empty PLY that fails silently in your analysis two steps later.

If you are going to look at the model rather than measure it, PLY is a poor destination: no materials, no textures, no structure, and a much larger file. Keep the glTF, or convert to OBJ if the viewer is old.

PLY earns its place when the next step is a geometry algorithm — decimation, remeshing, hole filling, alignment against a scan, a volume or surface-area measurement, a Hausdorff distance between two versions of the same part. For that work the flatness and the colourlessness are irrelevant, and having the mesh in the format the tool was written around is worth more than anything the conversion drops.