-
Notifications
You must be signed in to change notification settings - Fork 56
/
prepare_shapenet_dataset.py
193 lines (151 loc) · 8.42 KB
/
prepare_shapenet_dataset.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
import os
# Enable this when running on a computer without a screen:
# os.environ['PYOPENGL_PLATFORM'] = 'egl'
import trimesh
from tqdm import tqdm
import numpy as np
from util import ensure_directory
from multiprocessing import Pool
import traceback
from mesh_to_sdf import get_surface_point_cloud,scale_to_unit_cube, scale_to_unit_sphere, BadMeshException
DATASET_NAME = 'chairs'
DIRECTORY_MODELS = 'data/shapenet/03001627'
MODEL_EXTENSION = '.obj'
DIRECTORY_VOXELS = 'data/{:s}/voxels_{{:d}}/'.format(DATASET_NAME)
DIRECTORY_UNIFORM = 'data/{:s}/uniform/'.format(DATASET_NAME)
DIRECTORY_SURFACE = 'data/{:s}/surface/'.format(DATASET_NAME)
DIRECTORY_SDF_CLOUD = 'data/{:s}/cloud/'.format(DATASET_NAME)
DIRECTORY_BAD_MESHES = 'data/{:s}/bad_meshes/'.format(DATASET_NAME)
# Voxel resolutions to create.
# Set to [] if no voxels are needed.
# Set to [32] for for all models except for the progressively growing DeepSDF/Voxel GAN
VOXEL_RESOLUTIONS = [8, 16, 32, 64]
CREATE_SDF_CLOUDS = False # For DeepSDF autodecoder, contains uniformly and non-uniformly sampled points as proposed in the DeepSDF paper
CREATE_UNIFORM_AND_SURFACE = True # Uniformly sampled points for the Pointnet-based GAN and surface point clouds for the pointnet-based GAN with refinement
SDF_POINT_CLOUD_SIZE = 200000 # For DeepSDF point clouds (CREATE_SDF_CLOUDS)
POINT_CLOUD_SAMPLE_SIZE = 64**3 # For uniform and surface points (CREATE_UNIFORM_AND_SURFACE)
# Options for virtual scans used to generate SDFs
USE_DEPTH_BUFFER = True
SCAN_COUNT = 50
SCAN_RESOLUTION = 1024
def get_model_files():
for directory, _, files in os.walk(DIRECTORY_MODELS):
for filename in files:
if filename.endswith(MODEL_EXTENSION):
yield os.path.join(directory, filename)
def get_hash(filename):
return filename.split('/')[-3]
def get_voxel_filename(model_filename, resolution):
return os.path.join(DIRECTORY_VOXELS.format(resolution), get_hash(model_filename) + '.npy')
def get_uniform_filename(model_filename):
return os.path.join(DIRECTORY_UNIFORM, get_hash(model_filename) + '.npy')
def get_surface_filename(model_filename):
return os.path.join(DIRECTORY_SURFACE, get_hash(model_filename) + '.npy')
def get_sdf_cloud_filename(model_filename):
return os.path.join(DIRECTORY_SDF_CLOUD, get_hash(model_filename) + '.npy')
def get_bad_mesh_filename(model_filename):
return os.path.join(DIRECTORY_BAD_MESHES, get_hash(model_filename))
def mark_bad_mesh(model_filename):
filename = get_bad_mesh_filename(model_filename)
ensure_directory(os.path.dirname(filename))
open(filename, 'w').close()
def is_bad_mesh(model_filename):
return os.path.exists(get_bad_mesh_filename(model_filename))
def get_uniform_and_surface_points(surface_point_cloud, number_of_points = 200000):
unit_sphere_points = np.random.uniform(-1, 1, size=(number_of_points * 2, 3)).astype(np.float32)
unit_sphere_points = unit_sphere_points[np.linalg.norm(unit_sphere_points, axis=1) < 1]
uniform_points = unit_sphere_points[:number_of_points, :]
distances, indices = surface_point_cloud.kd_tree.query(uniform_points)
uniform_sdf = distances.astype(np.float32).reshape(-1) * -1
uniform_sdf[surface_point_cloud.is_outside(uniform_points)] *= -1
surface_points = surface_point_cloud.points[indices[:, 0], :]
near_surface_points = surface_points + np.random.normal(scale=0.0025, size=surface_points.shape).astype(np.float32)
near_surface_sdf = surface_point_cloud.get_sdf(near_surface_points, use_depth_buffer=USE_DEPTH_BUFFER)
model_size = np.count_nonzero(uniform_sdf < 0) / number_of_points
if model_size < 0.01:
raise BadMeshException()
return uniform_points, uniform_sdf, near_surface_points, near_surface_sdf
def process_model_file(filename):
try:
if is_bad_mesh(filename):
return
mesh = trimesh.load(filename)
voxel_filenames = [get_voxel_filename(filename, resolution) for resolution in VOXEL_RESOLUTIONS]
if not all(os.path.exists(f) for f in voxel_filenames):
mesh_unit_cube = scale_to_unit_cube(mesh)
surface_point_cloud = get_surface_point_cloud(mesh_unit_cube, bounding_radius=3**0.5, scan_count=SCAN_COUNT, scan_resolution=SCAN_RESOLUTION)
try:
for resolution in VOXEL_RESOLUTIONS:
voxels = surface_point_cloud.get_voxels(resolution, use_depth_buffer=USE_DEPTH_BUFFER, check_result=True)
np.save(get_voxel_filename(filename, resolution), voxels)
del voxels
except BadMeshException:
tqdm.write("Skipping bad mesh. ({:s})".format(get_hash(filename)))
mark_bad_mesh(filename)
return
del mesh_unit_cube, surface_point_cloud
create_uniform_and_surface = CREATE_UNIFORM_AND_SURFACE and (not os.path.exists(get_uniform_filename(filename)) or not os.path.exists(get_surface_filename(filename)))
create_sdf_clouds = CREATE_SDF_CLOUDS and not os.path.exists(get_sdf_cloud_filename(filename))
if create_uniform_and_surface or create_sdf_clouds:
mesh_unit_sphere = scale_to_unit_sphere(mesh)
surface_point_cloud = get_surface_point_cloud(mesh_unit_sphere, bounding_radius=1, scan_count=SCAN_COUNT, scan_resolution=SCAN_RESOLUTION)
try:
if create_uniform_and_surface:
uniform_points, uniform_sdf, near_surface_points, near_surface_sdf = get_uniform_and_surface_points(surface_point_cloud, number_of_points=POINT_CLOUD_SAMPLE_SIZE)
combined_uniform = np.concatenate((uniform_points, uniform_sdf[:, np.newaxis]), axis=1)
np.save(get_uniform_filename(filename), combined_uniform)
combined_surface = np.concatenate((near_surface_points, near_surface_sdf[:, np.newaxis]), axis=1)
np.save(get_surface_filename(filename), combined_surface)
if create_sdf_clouds:
sdf_points, sdf_values = surface_point_cloud.sample_sdf_near_surface(use_scans=True, sign_method='depth' if USE_DEPTH_BUFFER else 'normal', number_of_points=SDF_POINT_CLOUD_SIZE, min_size=0.015)
combined = np.concatenate((sdf_points, sdf_values[:, np.newaxis]), axis=1)
np.save(get_sdf_cloud_filename(filename), combined)
except BadMeshException:
tqdm.write("Skipping bad mesh. ({:s})".format(get_hash(filename)))
mark_bad_mesh(filename)
return
del mesh_unit_sphere, surface_point_cloud
except:
traceback.print_exc()
def process_model_files():
for res in VOXEL_RESOLUTIONS:
ensure_directory(DIRECTORY_VOXELS.format(res))
if CREATE_UNIFORM_AND_SURFACE:
ensure_directory(DIRECTORY_UNIFORM)
ensure_directory(DIRECTORY_SURFACE)
if CREATE_SDF_CLOUDS:
ensure_directory(DIRECTORY_SDF_CLOUD)
ensure_directory(DIRECTORY_BAD_MESHES)
files = list(get_model_files())
worker_count = os.cpu_count() // 2
print("Using {:d} processes.".format(worker_count))
pool = Pool(worker_count)
progress = tqdm(total=len(files))
def on_complete(*_):
progress.update()
for filename in files:
pool.apply_async(process_model_file, args=(filename,), callback=on_complete)
pool.close()
pool.join()
def combine_sdf_clouds():
import torch
print("Combining SDF point clouds...")
files = list(sorted(get_model_files()))
files = [f for f in files if os.path.exists(get_sdf_cloud_filename(f))]
N = len(files)
points = torch.zeros((N * SDF_POINT_CLOUD_SIZE, 3))
sdf = torch.zeros((N * SDF_POINT_CLOUD_SIZE))
position = 0
for file_name in tqdm(files):
numpy_array = np.load(get_sdf_cloud_filename(file_name))
points[position * SDF_POINT_CLOUD_SIZE:(position + 1) * SDF_POINT_CLOUD_SIZE, :] = torch.tensor(numpy_array[:, :3])
sdf[position * SDF_POINT_CLOUD_SIZE:(position + 1) * SDF_POINT_CLOUD_SIZE] = torch.tensor(numpy_array[:, 3])
del numpy_array
position += 1
print("Saving combined SDF clouds...")
torch.save(points, os.path.join('data', 'sdf_points.to'))
torch.save(sdf, os.path.join('data', 'sdf_values.to'))
if __name__ == '__main__':
process_model_files()
if CREATE_SDF_CLOUDS:
combine_sdf_clouds()