-
Notifications
You must be signed in to change notification settings - Fork 0
/
space_carving.py
254 lines (183 loc) · 7.4 KB
/
space_carving.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
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
import numpy as np
from matplotlib import pyplot as plt
from utils import load_cameras, load_point_cloud, project, create_silhouettes, create_gif
INSIDE, OUTSIDE, OVERLAP, NOT_SET = 0, 1, 2, 3
class Projection_Labler:
"""
Class to label a cube as inside, outside or overlapping silhuettes
of an object.
"""
def __init__(self, cameras):
self.N_CAMS, self.RMAX, self.CMAX = (37, 576, 720)
self.cameras = cameras
self.silhouettes = np.empty((self.N_CAMS, self.RMAX, self.CMAX), dtype=np.uint8)
with open("data/silhouettes.npy", "rb") as f:
self.silhouettes = np.load(f)
print(f"Silhouettes loaded")
self.silhouettes[self.silhouettes == 254] = 0
self.silhouettes[self.silhouettes == 255] = 1
def label(self, points):
assert(points.shape[0] == (3))
label = INSIDE
for j in range(36):
# Project corners onto the image plane of the silhuette
projected = np.round(project(points, self.cameras[j])).astype(int)
# Create a filled area from the resulting square
area = self.fill_area(projected)
# Count the overlap of the area with the silhuette
inside = np.count_nonzero(self.silhouettes[j][area] > 0)
count_area = np.count_nonzero(area)
# Adjust label accordingly
if inside == 0:
label = OUTSIDE
break
elif inside != count_area:
label = OVERLAP
return label
def fill_area(self, projected_points):
x_min = np.maximum(np.min(projected_points[0,:]), 0)
y_min = np.maximum(np.min(projected_points[1,:]), 0)
x_max = np.minimum(np.max(projected_points[0,:]), self.CMAX)
y_max = np.minimum(np.max(projected_points[1,:]), self.RMAX)
mask = np.zeros((self.RMAX, self.CMAX)).astype(int)
mask[y_min:y_max,x_min:x_max] = 1
return mask > 0
class Octree:
def __init__(self, point, size, labler, depth = 0, label=NOT_SET):
self.children = []
self.depth = depth
self.label = label
self.labler = labler
self.point = point
self.size = size
def subdivide(self):
if self.label == INSIDE:
return
x, y, z = self.point
# Split space into 8 octants
new_size = self.size / 2
p_new = np.empty((8,3))
p_new[0,:] = np.array([x,y,z])
p_new[1,:] = np.array([x+new_size,y,z])
p_new[2,:] = np.array([x,y+new_size,z])
p_new[3,:] = np.array([x,y,z+new_size])
p_new[4,:] = np.array([x+new_size,y+new_size,z])
p_new[5,:] = np.array([x+new_size,y,z+new_size])
p_new[6,:] = np.array([x,y+new_size,z+new_size])
p_new[7,:] = np.array([x+new_size,y+new_size,z+new_size])
for i in range(8):
p = p_new[i]
label = self.label
xi, yi, zi = p
# Create corners for the current point
corners = np.empty((8,3))
corners[0,:] = np.array([xi,yi,zi])
corners[1,:] = np.array([xi+new_size,yi,zi])
corners[2,:] = np.array([xi,yi+new_size,zi])
corners[3,:] = np.array([xi,yi,zi+new_size])
corners[4,:] = np.array([xi+new_size,yi+new_size,zi])
corners[5,:] = np.array([xi+new_size,yi,zi+new_size])
corners[6,:] = np.array([xi,yi+new_size,zi+new_size])
corners[7,:] = np.array([xi+new_size,yi+new_size,zi+new_size])
label = self.labler.label(corners.T)
# Dont create voxels outside the silhouettes
if label == OUTSIDE:
continue
vox = Octree(p, new_size, self.labler, depth=self.depth + 1, label=label)
self.children.append(vox)
def subdivide_voxel(root, depth):
if root.depth == depth:
return
else:
root.subdivide()
for vox in root.children:
subdivide_voxel(vox, depth)
def get_leaf_points(root, depth):
"""
Find "leaf" points of an octree at a given depth
"""
if depth != root.depth and len(root.children) != 0:
points = np.empty(3)
for child in root.children:
points = np.vstack((points, get_leaf_points(child, depth)))
points = points[1:, :]
return points
else:
return root.point
def create_octree(point_cloud, cameras) -> Octree:
x_min = np.min(point_cloud[:,0])
y_min = np.min(point_cloud[:,1])
z_min = np.min(point_cloud[:,2])
x_max = np.max(point_cloud[:,0])
y_max = np.max(point_cloud[:,1])
z_max = np.max(point_cloud[:,2])
start_point = np.floor(np.array([x_min, y_min, z_min])*100)/100
size = np.ceil(np.max([x_max-x_min, y_max-y_min, z_max-z_min])*100)/100
labler = Projection_Labler(cameras)
# Initialize Octree
root = Octree(start_point, size, labler)
return root
def visualize_octree(root, depth):
"""
Visualize 3D points with space carving. Create Octree representation subdivided to a choosen depth
### Parameters:
root: Octree root
depth: subdivide depth, level of detail
"""
points = get_leaf_points(root, depth).reshape((-1,3))
# Normalize cloud points between 0 - 1
points += -np.min(points, axis=0)
points /= np.max(points) if not np.max(points) == 0 else 1.0
# Numb voxels in each dimensio
n_voxels = np.power(2,depth)
# Scale to match number of voxels in each dim
points *= n_voxels - 1
points = np.round(points).astype(int)
# Prepare ind
vox = np.zeros((n_voxels, n_voxels, n_voxels))
# Fill voxels from the base points
vox[points[:,0], points[:,1], points[:,2]] = 1
_, (ax) = plt.subplots(1, 1, subplot_kw={'projection':'3d', 'aspect':'auto'})
ax.set_title(f"Octree generated voxels, depth: {depth}")
ax.voxels(vox)
ax.view_init(elev=10, azim=-45)
plt.savefig(fname=f"1_{depth}", dpi=200)
ax.view_init(elev=10, azim=128)
plt.savefig(f"2_{depth}", dpi=200)
ax.view_init(elev=10, azim=150)
plt.savefig(f"3_{depth}", dpi=200)
def visualize_pointcloud(point_cloud):
"""
Visualize 3D points.
### Parameters:
point_cloud: 3D points cloud shape (numb_points, 3)
"""
point_cloud += -np.min(point_cloud, axis=0)
point_cloud /= np.max(point_cloud)
fig, (ax) = plt.subplots(1, 1, subplot_kw={'projection':'3d', 'aspect':'auto'})
ax.set_title("Point cloud")
ax.scatter(point_cloud[:, 0], point_cloud[:, 1], point_cloud[:, 2], s=1, c='r', zorder=10)
ax.view_init(elev=10, azim=-45)
plt.savefig(fname=f"PC_1", dpi=200)
ax.view_init(elev=10, azim=128)
plt.savefig(fname=f"PC_2", dpi=200)
ax.view_init(elev=10, azim=150)
plt.savefig(fname=f"PC_3", dpi=200)
#plt.show()
if __name__ == "__main__":
# create_silhouettes()
# # Load point cloud
point_cloud = load_point_cloud("data/point_cloud.npy")
# visualize_pointcloud(point_cloud)
# # Load camera matrices
cameras = load_cameras("data/cameras.csv")
octree = create_octree(point_cloud, cameras)
subdivide_voxel(octree, 1)
visualize_octree(octree, depth=0)
visualize_octree(octree, depth=2)
# visualize_octree(octree, depth=3)
# visualize_octree(octree, depth=4)
# visualize_octree(octree, depth=5)
# visualize_octree(octree, depth=8)
# visualize_octree(octree, depth=7)
# visualize_octree(octree, depth=8)