Meshiki

A collection of unusual mesh processing algorithms.

Install

# from pypi, will build extension at first time
pip install meshiki

# locally
cd meshiki
pip install . 

Usage

Trigs-to-Quads

          Triangulate       We want this!
    Quads ----------> Trigs ------------> Quads
(obj, blender)   (glb, fbx, ...)

This algorithm is aimed for converting a triangulated quad-dominant mesh back to a mixed tri/quad mesh, with as many as possible reasonable quad faces. Our implementation is based on maximum weighted graph matching, and is usually better compared to the built-in tool (Edit Mode -> Face -> Tris to Quads) of blender.

from meshiki import Mesh

mesh = Mesh.load('mesh.glb', verbose=False)
mesh.quadrangulate()
mesh.export('mesh.obj') # must use obj for quad faces

Salient point sampling

This algorithm samples salient points from mesh surface as proposed in Dora.

from meshiki import Mesh, fps, load_mesh, triangulate

# load mesh
vertices, faces = load_mesh(mesh_path, clean=True)
# make sure it's pure-trig
faces = triangulate(faces)
mesh = Mesh(vertices, faces)
# sample 64K salient points
salient_points = mesh.salient_point_sample(64000, thresh_bihedral=30) # np.ndarray, [64000, 3]

We also implement uniform sampling and furthest point sampling:

# sample 128K uniform points
uniform_points = mesh.uniform_point_sample(128000) # np.ndarray, [128000, 3]
# use FPS to subsample 8K points from uniform points
fps_points = fps(uniform_points, N_FPS, backend='kdline') # np.ndarray, [8000, 3]

Acknowledgement

• QuickFPS and fpsample for fast furthest point sampling.
• PCG32.
• nanoflann.