PyNanoInstantMeshes

March 13, 2026 · View on GitHub

Python bindings for Instant Meshes, a tool for automatic remeshing of 3D surfaces.

Installation

pip install PyNanoInstantMeshes

Usage

import numpy as np
import pynanoinstantmeshes as PyNIM

# vertices: float32 numpy array of shape (N, 3)
# faces:    uint32  numpy array of shape (M, 3)
new_verts, new_faces = PyNIM.remesh(vertices, faces, vertex_count=1000)

Example with trimesh

import trimesh
import numpy as np
import pynanoinstantmeshes as PyNIM

# Load or create a mesh
mesh = trimesh.primitives.Capsule(radius=1, height=2).to_mesh()

vertices = np.asarray(mesh.vertices, dtype=np.float32)
faces    = np.asarray(mesh.faces,    dtype=np.uint32)

# Remesh to approximately 500 vertices
new_verts, new_faces = PyNIM.remesh(vertices, faces, vertex_count=500)

result = trimesh.Trimesh(vertices=new_verts, faces=new_faces)
result.show()

API Reference

`pynanoinstantmeshes.remesh`

pynanoinstantmeshes.remesh(
    verts,
    faces,
    vertex_count,
    rosy=4,
    posy=4,
    scale=-1.0,
    face_count=-1,
    creaseAngle=0.0,
    align_to_boundaries=False,
    extrinsic=True,
    smooth_iter=0,
    knn_points=1000,
    deterministic=False,
) -> tuple[np.ndarray, np.ndarray]

Remeshes the input mesh and returns the new mesh as a (vertices, faces) tuple.

Parameters

Parameter	Type	Default	Description
`verts`	`np.ndarray` (N×3, float32)	—	Input vertex positions.
`faces`	`np.ndarray` (M×3, uint32)	—	Input triangle faces. Pass an empty array to treat `verts` as a point cloud.
`vertex_count`	`int`	—	Target number of vertices in the output mesh. Mutually exclusive with `scale` and `face_count`; the first positive value among the three is used.
`rosy`	`int`	`4`	Rotational symmetry order of the orientation field (`2` or `4`).
`posy`	`int`	`4`	Positional symmetry order (`4` = quad output, `6` = triangle output).
`scale`	`float`	`-1.0`	Target edge length. When positive, overrides `vertex_count` and `face_count`.
`face_count`	`int`	`-1`	Target number of faces in the output mesh. Used when positive and `scale ≤ 0`.
`creaseAngle`	`float`	`0.0`	Dihedral-angle threshold (degrees) for crease detection. Edges whose dihedral angle exceeds this value are treated as sharp creases.
`align_to_boundaries`	`bool`	`False`	When `True`, aligns the orientation field to mesh boundaries.
`extrinsic`	`bool`	`True`	Use extrinsic (world-space) rather than intrinsic (surface-space) smoothness for the orientation field.
`smooth_iter`	`int`	`0`	Number of Laplacian smoothing iterations applied to the output mesh.
`knn_points`	`int`	`1000`	Number of nearest neighbours used when processing a point cloud (`faces` is empty).
`deterministic`	`bool`	`False`	Use a deterministic (single-threaded) algorithm for reproducible results.

Returns

A tuple (new_verts, new_faces) where:

new_verts – np.ndarray of shape (V, 3), float32 — output vertex positions.
new_faces – np.ndarray of shape (F, D), uint32 — output face indices, where D = 4 for quad meshes (posy=4) and D = 3 for triangle meshes (posy=6).