Python三维网格体素化实例_Python

python三维网格体素化

本文主要是实现将一个网格模型体素化，实现不同分辨率的体素化效果，并且可视化输出为obj文件！

首先利用trimesh对mesh进行采样，然后根据采样点得到各个体素点的占有值。

效果

通过调整分辨率以及采样率（当分辨率变高时建议适量提高采样率）得到以下的效果！

代码

#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
@author: matthieu zins
"""

import trimesh
import numpy as np
import os
import argparse

"""
    ======   voxelize the surface of a mesh   ======
"""



def create_if_needed(folder):
    if not os.path.isdir(folder):
        os.mkdir(folder)



parser = argparse.argumentparser(description='pass object name')
parser.add_argument('input_mesh', type=str)
parser.add_argument('--output_folder', type=str, default="")
parser.add_argument('--resolution', type=int, nargs=3, default=[50, 50, 50], 
                    help="resolution_x resolution_y resolution_z")
parser.add_argument('--sampling', type=int, default="100000",
                    help="number of points sampled on the mesh")
args = parser.parse_args()


input_mesh_filename = args.input_mesh
object_name = os.path.splitext(os.path.basename(input_mesh_filename))[0]
output_folder = args.output_folder
if len(output_folder) == 0: output_folder = object_name
res_x, res_y, res_z = args.resolution
sample_points_count = args.sampling

create_if_needed(output_folder)


mesh = trimesh.exchange.load.load(input_mesh_filename)


# uniform points sampling
pts, _ = trimesh.sample.sample_surface_even(mesh, sample_points_count )

# save sample points
sampled_points_mesh = trimesh.trimesh(vertices=pts)
sampled_points_mesh.export(os.path.join(output_folder, object_name + "_resampled_points.ply"))


# adjust the grid origin and voxels size
origin = pts.min(axis=0)
dimensions = pts.max(axis=0) - pts.min(axis=0)
scales = np.divide(dimensions, np.array([res_x-1, res_y-1, res_z-1]))
scale = np.max(scales)


# voxelize

pts -= origin
pts /= scale
pts_int = np.round(pts).astype(int)

grid = np.zeros((res_x, res_y, res_z), dtype=int)
goores_x = np.where(np.logical_and(pts_int[:, 0] >= 0, pts_int[:, 0] < res_x))[0]
goores_y = np.where(np.logical_and(pts_int[:, 1] >= 0, pts_int[:, 1] < res_y))[0]
goores_z = np.where(np.logical_and(pts_int[:, 2] >= 0, pts_int[:, 2] < res_z))[0]
goods = np.intersect1d(np.intersect1d(goores_x, goores_y), goores_z)
pts_int = pts_int[goods, :]
grid[pts_int[:, 0], pts_int[:, 1], pts_int[:, 2]] = 1




# save voxels
voxel_pts = np.array([[-0.5, 0.5, -0.5],
                      [0.5, 0.5, -0.5],
                      [0.5, 0.5, 0.5],
                      [-0.5, 0.5, 0.5],
                      [-0.5, -0.5, -0.5],
                      [0.5, -0.5, -0.5],
                      [0.5, -0.5, 0.5],
                      [-0.5, -0.5, 0.5]])
voxel_faces = np.array([[0, 1, 2, 3],
                        [1, 5, 6, 2],
                        [5, 4, 7, 6],
                        [4, 0, 3, 7],
                        [0, 4, 5, 1],
                        [7, 3, 2, 6]])

def get_voxel(i, j, k):
    global voxel_pts, voxel_faces
    v = np.array([i, j, k], dtype=float) * scale
    v += origin
    points = voxel_pts * scale + v
    return points, voxel_faces.copy()

points = []
faces = []
fi = 0
for i in range(res_x):
    for j in range(res_y):
        for k in range(res_z):
            if grid[i, j, k]:
                p, f = get_voxel(i, j, k)
                points.append(p)
                f += fi
                faces.append(f)
                fi += 8

points = np.vstack(points)
faces = np.vstack(faces)
# write obj mesh with quad faces
with open(os.path.join(output_folder, object_name + "_voxels.obj"), "w") as fout:
    for p in points:fout.write("v " + " ".join(map(str, p)) + "\n")
    for f in faces+1:fout.write("f " + " ".join(map(str, f)) + "\n")


print(object_name, "done.")

运行：

dependencies

numpy
trimesh

## usage

python voxelize_surface.py example/chair.obj --output_folder output --resolution 30 30 30 --sampling 10000

the optional parameters are：

output_folder (string): folder where the result is saved
resolution (list): the resolution of the grid [res_x, res_y, res_z]
sampling (int): number of points sampled on the mesh surface before voxelization
注：输入的类型可以时obj也可以是off以及ply格式！

注意：

若出现如下情况，可将采样点数（sampling）提高！

出现此种情况的原因是采样间隔太大，而体素尺寸太小（分辨太高），所以导致在有些体素的占有值进行判断的时候出现错误。

所以也可以通过降低分辨率来改善此种情况！