-
Notifications
You must be signed in to change notification settings - Fork 0
/
HW1.py
215 lines (173 loc) · 6.52 KB
/
HW1.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
import cv2
import numpy as np
import open3d as o3d
import matplotlib.pyplot as plt
import scipy
from sklearn.preprocessing import normalize
image_row = 120
image_col = 120
# visualizing the mask (size : "image width" * "image height")
def mask_visualization(M):
mask = np.copy(np.reshape(M, (image_row, image_col)))
plt.figure()
plt.imshow(mask, cmap='gray')
plt.title('Mask')
# visualizing the unit normal vector in RGB color space
# N is the normal map which contains the "unit normal vector" of all pixels (size : "image width" * "image height" * 3)
def normal_visualization(N):
# converting the array shape to (w*h) * 3 , every row is a normal vetor of one pixel
N_map = np.copy(np.reshape(N, (image_row, image_col, 3)))
# Rescale to [0,1] float number
N_map = (N_map + 1.0) / 2.0
plt.figure()
plt.imshow(N_map)
plt.title('Normal map')
# visualizing the depth on 2D image
# D is the depth map which contains "only the z value" of all pixels (size : "image width" * "image height")
def depth_visualization(D):
D_map = np.copy(np.reshape(D, (image_row,image_col)))
# D = np.uint8(D)
plt.figure()
plt.imshow(D_map)
plt.colorbar(label='Distance to Camera')
plt.title('Depth map')
plt.xlabel('X Pixel')
plt.ylabel('Y Pixel')
# convert depth map to point cloud and save it to ply file
# Z is the depth map which contains "only the z value" of all pixels (size : "image width" * "image height")
def save_ply(Z, filepath):
Z_map = np.reshape(Z, (image_row, image_col)).copy()
data = np.zeros((image_row * image_col,3),dtype=np.float32)
# let all point float on a base plane
baseline_val = np.min(Z_map)
Z_map[np.where(Z_map == 0)] = baseline_val
for i in range(image_row):
for j in range(image_col):
idx = i * image_col + j
data[idx][0] = j
data[idx][1] = i
data[idx][2] = Z_map[image_row - 1 - i][j]
# output to ply file
pcd = o3d.geometry.PointCloud()
pcd.points = o3d.utility.Vector3dVector(data)
o3d.io.write_point_cloud(filepath, pcd,write_ascii=True)
# show the result of saved ply file
def show_ply(filepath):
pcd = o3d.io.read_point_cloud(filepath)
o3d.visualization.draw_geometries([pcd])
# read the .bmp file
def read_bmp(filepath):
global image_row
global image_col
image = cv2.imread(filepath, cv2.IMREAD_GRAYSCALE)
image_row, image_col = image.shape
return image
def read_data(filepath):
L = []
I = []
num = 0
# Get the light source from LightSource.txt
with open(filepath + '/LightSource.txt', 'r') as f:
for line in f.readlines():
line = line.strip()
point = list(map(int, line[line.find('(') + 1: line.find(')')].split(',')))
L.append(np.array(point).astype(np.float32))
num += 1
# Get the "unit vector" of light source, dim of L: (m, 3), m is number of light source
L = normalize(L, axis = 1)
# Get image data
for i in range(1, num + 1):
pic = read_bmp(filepath + '/pic' + str(i) + '.bmp')
# Flatten image size from H * W to HW
I.append(pic.ravel())
# Change type to ndarray
I = np.asarray(I)
return I, L
def get_normal_vector(I, L):
# I = L * Kd * N, solve the equation to get Kd * N
KdN = np.linalg.solve(L.T @ L, L.T @ I)
KdN = normalize(KdN, axis = 0)
return KdN.T
def fill_value_into_matrix(M, v, s, N, mask):
nonzero_h, nonzero_w = np.where(mask!=0)
# Calculate the index number used in filling M
index_array = np.zeros((image_row, image_col)).astype(np.int16)
for i in range(s):
index_array[nonzero_h[i], nonzero_w[i]] = i
for i in range(s):
h = nonzero_h[i]
w = nonzero_w[i]
n_x = N[h, w, 0]
n_y = N[h, w, 1]
n_z = N[h, w, 2]
# z(x+1, y) - z(x, y) = - nx / nz
j = i*2
if mask[h, w+1]:
k = index_array[h, w+1]
M[j, i] = -1
M[j, k] = 1
v[j] = -n_x/n_z
elif mask[h, w-1]:
k = index_array[h, w-1]
M[j, k] = -1
M[j, i] = 1
v[j] = -n_x/n_z
# z(x, y+1) - z(x, y) = -ny / nz
j = i*2+1
if mask[h+1, w]:
k = index_array[h+1, w]
M[j, i] = 1
M[j, k] = -1
v[j] = -n_y/n_z
elif mask[h-1, w]:
k = index_array[h, w-1]
M[j, k] = 1
M[j, i] = -1
v[j] = -n_y/n_z
return M, v
def get_surface_reconstruction(z, mask, s, optimize):
nonzero_h, nonzero_w = np.where(mask!=0)
# Filter strange point in z
normalized_z = (z - np.mean(z)) / np.std(z)
outliner_idx = np.abs(normalized_z) > 2
z_max = np.max(z[~outliner_idx])
z_min = np.min(z[~outliner_idx])
Z = mask.astype(np.float32)
if optimize:
for i in range(s):
if z[i] > z_max:
Z[nonzero_h[i], nonzero_w[i]] = z_max
elif z[i] < z_min:
Z[nonzero_h[i], nonzero_w[i]] = z_min
else:
Z[nonzero_h[i], nonzero_w[i]] = z[i]
else:
for i in range(s):
Z[nonzero_h[i], nonzero_w[i]] = z[i]
return Z
def compute_depth(mask, N, optimize = True):
N = np.reshape(N, (image_row, image_col, 3))
# number of pixels of the object
s = np.size(np.where(mask != 0)[0])
# Mz = V
M = scipy.sparse.lil_matrix((2*s, s))
v = np.zeros((2*s, 1))
# Fill the value into M and v
M, v = fill_value_into_matrix(M, v, s, N, mask)
# M.T * M * z = M.T * v
z = scipy.sparse.linalg.spsolve(M.T @ M, M.T @ v)
return get_surface_reconstruction(z, mask, s, optimize)
if __name__ == '__main__':
objects = ['star', 'bunny', 'venus']
for object in objects:
filepath = 'test/' + object
I, L = read_data(filepath)
N = get_normal_vector(I, L)
normal_visualization(N)
mask = read_bmp(filepath + '/pic1.bmp')
Z = compute_depth(mask, N, optimize = True)
depth_visualization(Z)
save_ply(Z, filepath + '/' + object + '.ply')
show_ply(filepath + '/' + object + '.ply')
# showing the windows of all visualization function
plt.show()