3D моделирование в Python

• Docker для исполнения примеров Pymesh:

curl -fsSL https://get.docker.com -o get-docker.sh

sudo sh get-docker.sh

• Компилятор g++ для PyTorch3d:

sudo apt update

sudo apt install g++

• Клонирование репозитория и установка библиотек:

git clone https://github.com/format37/python3d.git

cd python3d

pip install -r requirements.txt

pip install "git+https://github.com/facebookresearch/pytorch3d.git"

• Установка OpenScad:

sudo apt-get install openscad

Что такое jupyter-ноутбук и зачем он нужен

Договоимся что длина стороны грани фрактала - единица. Глубина фрактала это, грубо говоря, количество уникальных размеров отверстий. Длина вокселя зависит от глубины фрактала, она делится на 3 с каждым новым уровнем глубины.

Найдем сторону вокселя на глубинах 1 и 2. Упростим задачу, свернув фрактал с 3-х до 1-мерного случая:

Если фрактал второго уровня, длина стороны куба составит 1/(3**2) или 1/9. Составим набор кубов так, чтобы своим расположением они заполнили исходный куб. Получится воксельный куб. Вычислим области отверстий. Исключим воксели, которые входят в области отверстий. В завершение, объединим оставшиеся воксели в один объект и сохраним.

from stl import mesh
import numpy as np
import datetime

time_start = datetime.datetime.now()


def create_voxel(side, tx, ty, tz):
    vertices = np.array([
       [tx + 0,     ty + 0,      tz + 0],
       [tx + side,  ty + 0,      tz + 0],
       [tx + side,  ty + side,   tz + 0],
       [tx + 0,     ty + side,   tz + 0],
       [tx + 0,     ty + 0,      tz + side],
       [tx + side,  ty + 0,      tz + side],
       [tx + side,  ty + side,   tz + side],
       [tx + 0,     ty + side,   tz + side]
    ])

    faces = np.array([
        [3, 2, 1],
        [3, 1, 0],
        [0, 1, 5],#
        [5, 4, 0],
        [7, 3, 0],
        [0, 4, 7],#
        [1, 2, 6],#
        [6, 5, 1],
        [2, 3, 6],#
        [3, 7, 6],#
        [4, 5, 6],#
        [6, 7, 4]
     ])
    voxel = mesh.Mesh(np.zeros(faces.shape[0], dtype=mesh.Mesh.dtype))
    for i, f in enumerate(faces):
        for j in range(3):
            voxel.vectors[i][j] = vertices[f[j],:]
    return voxel
  
  
def voxel_in_hole(holes, x, y):
    comp_error = 0
    for hole in holes:
        if  x + comp_error >= hole['x'][0] and x - comp_error < hole['x'][1] and \
            y + comp_error >= hole['y'][0] and y - comp_error < hole['y'][1]:
            return True
    return False
  
  
depth = 3
side = 1/(3**depth)

# holes
holes = []
side = 1
round_factor = 14
for d in range(1, depth+1):
    side /= 3
    for x in range(1,3**d,3):
        for y in range(1,3**d,3):           
            holes.append({
                'x':[round(side*x,round_factor), round(side*x+side,round_factor)],
                'y':[round(side*y,round_factor), round(side*y+side,round_factor)]
            })
            
sponge = None
for x in range(3**depth):
    for y in range(3**depth):
        for z in range(3**depth):
            rx = round(x/(3**depth),round_factor)
            ry = round(y/(3**depth),round_factor)
            rz = round(z/(3**depth),round_factor)
            if  voxel_in_hole(holes, ry, rz) or \
                voxel_in_hole(holes, rx, ry) or \
                voxel_in_hole(holes, rx, rz):
                continue
            if sponge is None:
                sponge = create_voxel(side, x/(3**depth), y/(3**depth), z/(3**depth))
            else:
                new_voxel = create_voxel(side, x/(3**depth), y/(3**depth), z/(3**depth))
                sponge = mesh.Mesh(np.concatenate([sponge.data, new_voxel.data]))
                
sponge.save('numpy_stl_example_04.stl')


time_end = datetime.datetime.now()
print('spent', (time_end - time_start).seconds, 'seconds')

from myplot import plot_mesh
from stl import mesh
import datetime
import imageio
import math
import os

sponge = mesh.Mesh.from_file('numpy_stl_example_04.stl')

gif_half_size = 14
filenames = []
imageio_images = []
for i in range(gif_half_size):
    sponge.rotate([1, 0.0, 1], math.radians(1))
    filename = 'solidpython_example_03_'+str(i)+'.png'
    filenames.append(filename)
    plot_mesh(sponge, filename = filename)
    print(datetime.datetime.now(), filename)
    
print('forward rendering..')
for i in range(gif_half_size):
    print(datetime.datetime.now(), filenames[i])
    imageio_images.append(imageio.imread(filenames[i]))
    i-=1
    
print('reverse rendering..')
i = gif_half_size-1
while i>0:
    print(datetime.datetime.now(), filenames[i])
    imageio_images.append(imageio.imread(filenames[i]))
    i-=1
    
# save gif
imageio.mimsave('solidpython_example_03.gif', imageio_images)
    
# remove images
for filename in filenames:
    os.remove(filename)
    
print('job complete!')

К сожалению, библиотека PyMesh не установилась у меня ни через менеджер пакетов PIP (несмотря на упоминание этого способа в документации), ни через Anaconda. Есть два способа установки.

1. Следуя инструкции, собрать из исходников.

2. Используя Docker контейнер. Выберем этот вариант, как более интересный. Запуск контейнера будет инициирован с параметрами. При помощи параметров запуска смонтируем в контейнер папку скриптов. Передадим необходимые параметры в скрипт. После завершения работы скрипта, контейнер будет удален. Если Docker вы уже установили следуя инструкции в начале статьи, более ничего устанавливать не нужно.

Если Docker вам не подходит, скомпилируйте PyMesh, следуя инструкции в документации. Такой вариант тоже будет работать.

• Запускается контейнер pymesh. При первом запуске будет загружен образ, это займёт некоторое время.

• Папка pymesh_examples монтируется в контейнер

• В контейнере запускается python скрипт /pymesh_examples/pymesh_example_01.py

• Импортируется библиотека pymesh

• функция generate_box_mesh генерирует куб на основании двух противоположных вершин в точках [0,0,0] и [1,1,1]

• функция save_mesh сохраняет объект в stl файл.

• После исполнения, в папке pymesh_examples появляется файл pymesh_example_01.stl

import pymesh
import numpy as np
import math

def mesh_trans(mesh, x, y, z):
   return pymesh.form_mesh(mesh.vertices + [[x, y, z]], mesh.faces)

def mesh_rotate(mesh, x, y, z, angle):
   rot = pymesh.Quaternion.fromAxisAngle(np.array([x, y, z]), math.pi*2*angle/360)
   return pymesh.form_mesh(np.dot(rot.to_matrix(), mesh.vertices.T).T, mesh.faces)

sponge = pymesh.generate_box_mesh([0,0,0], [1,1,1])
bool_box = pymesh.generate_box_mesh([0.4,0.4,-0.1], [0.6,0.6,1.1])
sponge = pymesh.boolean(sponge, bool_box, operation='difference', engine="igl")

bool_box = mesh_rotate(mesh = bool_box, x = 1, y = 0, z = 0, angle = 90)
bool_box = mesh_trans(mesh = bool_box, x = 0, y = 1, z = 0)
sponge = pymesh.boolean(sponge, bool_box, operation='difference', engine="igl")

filename = "/pymesh_examples/pymesh_example_03.stl"
pymesh.save_mesh(filename, sponge, ascii=False)

import numpy as np
import datetime
import pymesh
import math
import sys

time_start = datetime.datetime.now()

depth = int(sys.argv[1])

def mesh_trans(mesh, x, y, z):
   return pymesh.form_mesh(mesh.vertices + [[x, y, z]], mesh.faces)

def mesh_rotate(mesh, x, y, z, angle):
   rot = pymesh.Quaternion.fromAxisAngle(np.array([x, y, z]), math.pi*2*angle/360)
   return pymesh.form_mesh(np.dot(rot.to_matrix(), mesh.vertices.T).T, mesh.faces)

def menger_sponge(depth):
   mesh_fractal = pymesh.generate_box_mesh([0,0,0], [1,1,1])
   z = [-0.1,1.1]
   side = 1
   for d in range(1, depth+1):
      side /= 3
      for x in range(1,3**d,3):
         for y in range(1,3**d,3):
               print(datetime.datetime.now(), d, x, y, '/', depth, ':', 3**d)
               box_a = pymesh.generate_box_mesh([side*x,side*y,z[0]], [side*x+side,side*y+side,z[1]])
            
               box_b = mesh_rotate(mesh = box_a, x = 1, y = 0, z = 0, angle = 90)
               box_b = mesh_trans(mesh = box_b, x = 0, y = 1, z = 0)

               box_c = mesh_rotate(mesh = box_a, x = 0, y = 1, z = 0, angle = 90)
               box_c = mesh_trans(mesh = box_c, x = 0, y = 0, z = 1)

               mesh_fractal = pymesh.boolean(mesh_fractal, box_a, operation='difference', engine="igl")
               mesh_fractal = pymesh.boolean(mesh_fractal, box_b, operation='difference', engine="igl")
               mesh_fractal = pymesh.boolean(mesh_fractal, box_c, operation='difference', engine="igl")           

   return mesh_fractal

mesh_fractal = menger_sponge(depth)

pymesh.save_mesh("/pymesh_examples/pymesh_example_04_"+str(depth)+".stl", mesh_fractal, ascii=False)

time_end = datetime.datetime.now()
print('spent', (time_end - time_start).seconds, 'seconds')

Построение займет около 4-х часов, а размер файла составит 73 мб:

import pymesh
from myplot_vf import plot_vf
import imageio
import os
import datetime

gif_half_size = 16

sponge = pymesh.meshio.load_mesh('/pymesh_examples/pymesh_example_04_3.stl')

filenames = []
imageio_images = []

# rendering
for i in range(gif_half_size):
    vertices = sponge.vertices
    faces = sponge.faces
    filename = '/pymesh_examples/pymesh_example_05_3_'+str(i)+'.png'
    print(filename)
    filenames.append(filename)    
    plot_vf(vertices, faces, cam_x=i, cam_y=i, filename = filename)

print('forward rendering..')
for i in range(gif_half_size):
    print(datetime.datetime.now(), filenames[i])
    imageio_images.append(imageio.imread(filenames[i]))
    i-=1

print('reverse rendering..')
i = gif_half_size-1
while i>0:
    print(datetime.datetime.now(), filenames[i])
    imageio_images.append(imageio.imread(filenames[i]))
    i-=1

# save gif
imageio.mimsave('/pymesh_examples/pymesh_example_05_3.gif', imageio_images)

# remove images
for filename in filenames:
    os.remove(filename)

import torch
from pytorch3d.io import save_obj
from scipy import spatial
import numpy as np

if torch.cuda.is_available():
    device = torch.device("cuda:0")
    torch.cuda.set_device(device)
else:
    device = torch.device("cpu")

with_grad = False

verts = torch.tensor(
    [
        [-3, -3, 0],
        [+3, -3, 0],
        [+3, +3, 0],
        [-3, +3, 0],
        [+0, +0, +3]
    ],
    device=device,
    dtype=torch.float32,
    requires_grad=with_grad,
)

hull = spatial.ConvexHull(verts.cpu().numpy())
faces_data = hull.simplices

faces = torch.tensor(
    faces_data,
    device=device,
    dtype=torch.int64,
)

save_obj('pytorch3d_example_01.obj', verts, faces)

import torch
from pytorch3d.io import IO
from pytorch3d.structures.meshes import Meshes
from pytorch3d.structures import join_meshes_as_scene
from pytorch3d.transforms import RotateAxisAngle
from pytorch3d.transforms import Translate
from scipy import spatial
import numpy as np

if torch.cuda.is_available():
   device = torch.device("cuda:0")
   torch.cuda.set_device(device)
else:
   device = torch.device("cpu")

with_grad = False

host_verts = [
       [-3, -3, 0],
       [+3, -3, 0],
       [+3, +3, 0],
       [-3, +3, 0],
       [+0, +0, +3]
   ]

verts = torch.tensor(
   host_verts,
   device=device,
   dtype=torch.float32,
   requires_grad=with_grad,
)

hull = spatial.ConvexHull(host_verts)
faces = torch.tensor(
   hull.simplices,
   device=device,
   dtype=torch.int64,
)

mesh_a = Meshes(verts=[verts], faces=[faces])

rot = RotateAxisAngle(180,'X', device=device)
verts_b = rot.transform_points(mesh_a.verts_list()[0])

trans = Translate(0,0,5, device=device)
verts_b = trans.transform_points(verts_b)

mesh_b = Meshes(verts=[verts_b], faces=[faces])

mesh = join_meshes_as_scene([mesh_a, mesh_b])
IO().save_mesh(mesh, 'pytorch3d_example_02.obj')

import torch
from pytorch3d.io import IO
from pytorch3d.structures.meshes import Meshes
from pytorch3d.structures import join_meshes_as_scene
from pytorch3d.transforms import Translate
from scipy import spatial
import numpy as np
import sys
import datetime

batch_size = 10

time_start = datetime.datetime.now()

def log(message):
   print(datetime.datetime.now().strftime('%Y.%m.%d %H:%M:%S'), message)

if torch.cuda.is_available():
   device = torch.device("cuda:0")
   torch.cuda.set_device(device)
else:
   device = torch.device("cpu")

depth = 3
side = 1/(3**depth)
with_grad = False

# vertices
vertices_base = [
       [0,     0,      0],
       [side,  0,      0],
       [side,  side,   0],
       [0,     side,   0],
       [0,     0,      side],
       [side,  0,      side],
       [side,  side,   side],
       [0,     side,   side]
   ]
vertices = torch.tensor(
   vertices_base,
   device=device,
   dtype=torch.float32,
   requires_grad=with_grad,
)

# faces
#hull = spatial.ConvexHull(vertices_base)
faces_data = [
    [3, 2, 1],
    [3, 1, 0],
    [0, 1, 5],#
    [5, 4, 0],
    [7, 3, 0],
    [0, 4, 7],#
    [1, 2, 6],#
    [6, 5, 1],
    [2, 3, 6],#
    [3, 7, 6],#
    [4, 5, 6],#
    [6, 7, 4]
 ]
faces = torch.tensor(
   faces_data,
   device=device,
   dtype=torch.int64,
)

# holes
holes = []
side = 1
round_factor = 14
for d in range(1, depth+1):
   side /= 3
   for x in range(1,3**d,3):
       for y in range(1,3**d,3):           
           holes.append({
               'x':[round(side*x,round_factor), round(side*x+side,round_factor)],
               'y':[round(side*y,round_factor), round(side*y+side,round_factor)]
               })

def voxel_in_hole(holes, x, y):
   comp_error = 0
   for hole in holes:
       if  x + comp_error >= hole['x'][0] and x - comp_error < hole['x'][1] and \
           y + comp_error >= hole['y'][0] and y - comp_error < hole['y'][1]:
           return True
   return False

# voxels
voxels = []
batch = 0
mesh = None
for x in range(3**depth):
   log('x '+ str(x)+' in '+str(3**depth))
   for y in range(3**depth):
       for z in range(3**depth):
           if  voxel_in_hole(holes, round(y/(3**depth),round_factor), round(z/(3**depth),round_factor)) or \
               voxel_in_hole(holes, round(x/(3**depth),round_factor), round(y/(3**depth),round_factor)) or \
               voxel_in_hole(holes, round(x/(3**depth),round_factor), round(z/(3**depth),round_factor)):
               continue
           voxel_translate = Translate(x/(3**depth), y/(3**depth), z/(3**depth), device=device)
           voxel_vertices_translated = voxel_translate.transform_points(vertices)
           voxels.append(Meshes(verts=[voxel_vertices_translated], faces=[faces]))
   batch += 1
   if batch > batch_size:
       if not mesh is None:
           log('append mesh')
           voxels.append(mesh)
       log('join')
       mesh = join_meshes_as_scene(voxels)
       voxels = []
       batch = 0


log('join')
if len(voxels):
   if not mesh is None:
       voxels.append(mesh)
   mesh = join_meshes_as_scene(voxels)
log('save')
IO().save_mesh(mesh, 'pytorch3d_example_03.obj')
log('end')

time_end = datetime.datetime.now()
print('spent', (time_end - time_start).seconds, 'seconds')

В этом скрипте объявляем вершины минимального для указанной глубины вокселя. По знакомому из прошлого примера алгоритму вычисляем координаты отверстий в двух измерениях. Заполняем первичный куб вокселями, которые не попадают в пределы отверстий.

import torch
from pytorch3d.structures.meshes import Meshes
from pytorch3d.io import IO, load_obj
from pytorch3d.renderer import (
    FoVPerspectiveCameras, look_at_view_transform,
    RasterizationSettings, BlendParams,
    MeshRenderer, MeshRasterizer, HardPhongShader, TexturesVertex
)
import matplotlib.pyplot as plt
import imageio
import os

gif_half_size = 30

if torch.cuda.is_available():
   device = torch.device("cuda:0")
   torch.cuda.set_device(device)
else:
   device = torch.device("cpu")

# Load object
verts, faces, aux = load_obj('pytorch3d_example_03.obj')

# init texture
verts_list = []
faces_list = []
texts_list = []

verts = torch.as_tensor(verts, dtype=torch.float32, device=device)
faces = torch.as_tensor(faces.verts_idx, dtype=torch.float32, device=device)
texts = torch.rand((len(verts), 3), dtype=torch.float32, device=device)
verts_list.append(verts)
faces_list.append(faces)
texts_list.append(texts)

# create textures
tex = TexturesVertex(texts_list)

# Initialise the mesh with textures
mesh = Meshes(verts=[verts], faces=[faces], textures=tex)[0]

filenames = []
imageio_images = []

# rendering
for i in range(gif_half_size):

    filename = 'pytorch3d_example_'+str(i)+'.png'
    print(filename)
    filenames.append(filename)

    # Select the viewpoint using spherical angles  
    distance = 3.5   # distance from camera to the object
    elevation = 0.0 + i   # angle of elevation in degrees
    azimuth = 0.0 + i  # No rotation so the camera is positioned on the +Z axis.

    # Get the position of the camera based on the spherical angles
    R, T = look_at_view_transform(distance, elevation, azimuth, device=device)

    # Initialize an OpenGL perspective camera.
    cameras = FoVPerspectiveCameras(device=device, R=R, T=T)

    # Define the settings for rasterization and shading. Here we set the output image to be of size
    # 512x512. As we are rendering images for visualization purposes only we will set faces_per_pixel=1
    # and blur_radius=0.0. Refer to rasterize_meshes.py for explanations of these parameters.
    raster_settings = RasterizationSettings(
        image_size=512,
        blur_radius=0.0,
        faces_per_pixel=1,
    )

    # Create a phong renderer by composing a rasterizer and a shader. Here we can use a predefined
    # PhongShader, passing in the device on which to initialize the default parameters
    renderer = MeshRenderer(
        rasterizer=MeshRasterizer(cameras=cameras, raster_settings=raster_settings),
        shader=HardPhongShader(device=device, cameras=cameras)
    )

    images = renderer(mesh)
    plt.figure(figsize=(10, 10))
    plt.imshow(images[0, ..., :3].cpu().numpy())
    plt.savefig(filename)
    plt.axis("off")
    imageio_images.append(imageio.imread(filename))
    

# reverse rendering
i = gif_half_size-1
while i>0:
    print(filenames[i])
    imageio_images.append(imageio.imread(filenames[i]))
    i-=1

# save gif
imageio.mimsave('pytorch3d_example_03.gif', imageio_images)

# remove images
for filename in filenames:
    os.remove(filename)

from solid import *
from IPython.display import Image
import datetime

depth = 3
sponge = cube([1,1,1], center = True)
z = 1.1
side = 1
for d in range(1, depth+1):
    side /= 3
    for x in range(1,3**d,3):
        for y in range(1,3**d,3):
            box_a = cube([side,side,z], center = True)
            box_a = translate([side*x-0.5+side/2, side*y-0.5+side/2, 0]) (box_a)            
            box_b = rotate([90,0,0]) (box_a)
            box_c = rotate([0,90,0]) (box_a)
            sponge -= box_a
            sponge -= box_b
            sponge -= box_c
path = scad_render_to_file(sponge, 'solidpython_example_03.scad')

# export to stl
!openscad solidpython_example_03.scad -o solidpython_example_03.stl

time_end = datetime.datetime.now()
print('spent', (time_end - time_start).seconds, 'seconds')

from solid import *
import subprocess
import datetime
import imageio
import os

gif_half_size = 30
filenames = []
imageio_images = []
for i in range(gif_half_size):
    filename = 'solidpython_example_03_'+str(i)+'.png'
    filenames.append(filename)
    # export to png with camera params
    # camera parameters when exporting png: 
    # =translate_x,y,z,rot_x,y,z,dist or 
    # =eye_x,y,z,center_x,y,z
    rot_x = str(i-20)
    rot_y = str(i-20)
    rot_z = str(i-20)
    dist = str(5+i/10)
    cmd = "openscad solidpython_example_03.scad --camera '0,0,0,"+\
        rot_x+","+rot_y+","+rot_z+","+dist+"' -o "+filename
    MyOut = subprocess.Popen(
        cmd, 
        shell=True, 
        stdout=subprocess.PIPE, 
        stderr=subprocess.PIPE, 
        stdin=subprocess.PIPE
    )
    stdout,stderr = MyOut.communicate()
    imageio_images.append(imageio.imread(filename))
    print(datetime.datetime.now(), filename)    
    print('stdout', stdout)
    print('stderr', stderr)
    
print('reverse rendering..')
i = gif_half_size-1
while i>0:
    print(datetime.datetime.now(), filenames[i])
    imageio_images.append(imageio.imread(filenames[i]))
    i-=1
    
# save gif
imageio.mimsave('solidpython_example_03.gif', imageio_images)
    
# remove images
for filename in filenames:
    os.remove(filename)
    
print('job complete!')

import solid as scad
import subprocess
import os

# sponge
depth = 3
sponge = scad.cube([1,1,1], center = True)
z = 1.1
side = 1
for d in range(1, depth+1):
    side /= 3
    for x in range(1,3**d,3):
        for y in range(1,3**d,3):
            box_a = scad.cube([side,side,z], center = True)
            box_a = scad.translate([side*x-0.5+side/2, side*y-0.5+side/2, 0]) (box_a)            
            box_b = scad.rotate([90,0,0]) (box_a)
            box_c = scad.rotate([0,90,0]) (box_a)
            sponge -= box_a
            sponge -= box_b
            sponge -= box_c

sponge = scad.translate([3.3, 0.5, 0.5]) (sponge)

# text
text = scad.text('Pyth   n 3d', size = 1)
text_gray = scad.linear_extrude(height = 0.9) (text)
text_gray = scad.color('gray', 0.5) (text_gray)

text_white = scad.linear_extrude(height = 0.1) (text)
text_white = scad.color('white', 0.8) (text_white)
text_white = scad.translate([0, 0, 0.9]) (text_white)

text_blue = scad.linear_extrude(height = 0.1) (text)
text_blue = scad.color('blue', 0.2) (text_blue)
text_blue = scad.translate([0, 0, 0.9]) (text_blue)

# merge
objects = text_gray + text_white + text_blue
for i in range(0,10):
    sponge_entity = scad.color('magenta', 0.5/(i**2+1)) (sponge)
    sponge_entity = sponge_entity + scad.color('blue', 0.5/(i**2+1)) (sponge)
    objects += scad.translate([0, 0, i*2]) (sponge_entity)
    if i>0:
        objects += scad.translate([0, 0, -i*2]) (sponge_entity)
        objects += scad.translate([0, i*2, 0]) (sponge_entity)
        objects += scad.translate([0, -i*2, 0]) (sponge_entity)
        
# render
filename = 'solidpython_example_05'
render = scad.scad_render_to_file(objects, filename+'.scad')
# camera parameters when exporting png: 
#                                 =translate_x,y,z,rot_x,y,z,dist or 
#                                 =eye_x,y,z,center_x,y,z

#  =colorscheme: *Cornfield | Metallic | Sunset | 
#                                 Starnight | BeforeDawn | Nature | DeepOcean | 
#                                 Solarized | Tomorrow | Tomorrow 2 | Tomorrow 
#                                 Night | Monotone
cmd = "openscad "+filename+".scad --camera '3,2,2,20,5,-70,30' --imgsize=10000,3000 --colorscheme='Tomorrow' -o "+filename+".png"
MyOut = subprocess.Popen(
    cmd, 
    shell=True, 
    stdout=subprocess.PIPE, 
    stderr=subprocess.PIPE, 
    stdin=subprocess.PIPE
)
MyOut.stdout, MyOut.stderr

• Ubuntu 20.04.2 LTS

• Python 3.7.3

• Intel(R) Core(TM) i3-7350K CPU @ 4.20GHz

• RAM 35.2Gb

• GPU GeForce GTX 1080 Ti 11175 MB

• 3D ML. Часть 1: формы представления 3D-данных

• 3D ML. Часть 2: функции потерь в задачах 3D ML

• 3D ML. Часть 3: датасеты и фреймворки в 3D ML

• 3D ML. Часть 4: дифференциальный рендеринг

• 3D ML. Часть 5: Свертки на графах

• 3D ML. Часть 6: Обзор алгоритмов семантической сегментации облака точек

• PyTorch3D : RoboLove

• trimesh - Страница проекта

• Работа с 3D моделями в Python с использованием библиотеки OpenMesh

• openscadpy - Страница проекта

• Система скриптового 3д моделирования ZenCad

• zenscad - Документация

• Hands-on Guide to PyTorch 3D – A Library for Deep Learning with 3D Data

• Scripting the Unreal Editor using Python

• Nvidia kaolin

• Guide To NVIDIA’s Kaolin: A 3D Deep Learning Library

• 3D Рендеринг в matplotlib