-
Notifications
You must be signed in to change notification settings - Fork 9
/
captured_data.py
167 lines (135 loc) · 6.06 KB
/
captured_data.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
import torch
import numpy as np
import cv2
import h5py
from tqdm import trange
import imageio
import config
Float = torch.float64
device='cuda'
def process_mask(M):
if M.max() == 255: M //= 255
assert M.max() == 1
dist= (cv2.distanceTransform(M, cv2.DIST_L2, 0)-0).clip(0,1)\
- (cv2.distanceTransform(1-M, cv2.DIST_L2, 0)-1).clip(0,1) #[-1,+1]
mask = (dist + 1) / 2 #[0,1]
mask[-1] = 0.5
return mask
def generate_ray(resy, resx, K_inverse, R_inverse):
K_inverse = torch.tensor(K_inverse, device=device, dtype=Float)
R_inverse = torch.tensor(R_inverse, device=device, dtype=Float)
y_range = torch.arange(0, resy, device=device, dtype=Float)
x_range = torch.arange(0, resx, device=device, dtype=Float)
pixely, pixelx = torch.meshgrid(y_range, x_range)
pixelz = torch.ones_like(pixely)
pixel = torch.stack([pixelx, pixely, pixelz], dim=2).view([-1,3])
pixel_p = K_inverse @ pixel.T
pixel_world_p = R_inverse[:3,:3] @ pixel_p + R_inverse[:3, 3:4]
ray_origin = R_inverse[:3, 3:4] #[3x1]
ray_dir = pixel_world_p - ray_origin
ray_dir = ray_dir.T #[nx3]
ray_dir = ray_dir/ray_dir.norm(dim=1,keepdim=True)
ray_origin = ray_origin.T.expand_as(ray_dir)
return ray_origin, ray_dir
class Data:
def get_view(self, V_index):
screen_pixel, valid, mask, origin, ray_dir, camera_M = self.Views[V_index]
R, K, R_inverse, K_inverse = camera_M
screen_pixel = screen_pixel.to(device)
valid = valid .to(device)
mask = mask .to(device)
origin = origin .to(device)
ray_dir = ray_dir .to(device)
R_inverse = R_inverse .to(device)
K_inverse = K_inverse .to(device)
R = R .to(device)
K = K .to(device)
camera_M = (R, K, R_inverse, K_inverse)
return screen_pixel, valid, mask, origin, ray_dir, camera_M
def ray_view_generator(self):
index = list(np.arange(0, 72, 72//self.num_view))
# mouse debug
if self.name == 'mouse':
index = list(np.arange(-5, 10))
index = index + list(np.arange(22,40))
print('num_view ray', len(index))
while True:
np.random.shuffle(index)
for i in index: yield i % 72
def silh_view_generator(self):
index = list(np.arange(72))
print('num_view silh', len(index))
while True:
np.random.shuffle(index)
for i in index: yield i % 72
class Data_Pointgray(Data):
'''
data captured by camera pointgray
'''
def __init__(self, HyperParams):
self.resy=960
self.resx=1280
self.num_view = HyperParams['num_view']
self.name = HyperParams['name']
h5data = h5py.File(f'{config.data_path}{self.name}.h5','r')
self.Views = []
print('loading data..............')
for i in trange(72):
R = h5data['cam_proj'][i]
K = h5data['cam_k'][:]
R_inverse = np.linalg.inv(R)
K_inverse = np.linalg.inv(K)
screen_pixel = h5data['screen_position'][i]
target = screen_pixel
mask = h5data['mask'][i]
valid = screen_pixel[:,0] != 0
ray_origin = h5data['ray_origin'][i]
ray_dir = h5data['ray_dir'][i]
mask = process_mask(mask)
target = torch.tensor(target , dtype = Float).pin_memory()
valid = torch.tensor(valid , dtype = bool ).pin_memory()
mask = torch.tensor(mask , dtype = Float).pin_memory()
ray_origin = torch.tensor(ray_origin , dtype = Float).pin_memory()
ray_dir = torch.tensor(ray_dir , dtype = Float).pin_memory()
R_inverse = torch.tensor(R_inverse , dtype = Float).pin_memory()
K_inverse = torch.tensor(K_inverse , dtype = Float).pin_memory()
R = torch.tensor(R , dtype = Float).pin_memory()
K = torch.tensor(K , dtype = Float).pin_memory()
camera_M = (R, K, R_inverse, K_inverse)
self.Views.append((target, valid, mask, ray_origin, ray_dir, camera_M))
h5data.close()
class Data_Redmi(Data):
'''
data captured by my cellphone Redmi
'''
def __init__(self, HyperParams):
self.resy=1080
self.resx=1920
self.num_view = HyperParams['num_view']
self.name = HyperParams['name']
h5data = h5py.File(f'{config.data_path}{self.name}.h5','r')
self.Views = []
print('loading data..............')
for i in trange(72):
R = h5data['cam_proj'][i]
K = h5data['cam_k'][:]
R_inverse = np.linalg.inv(R)
K_inverse = np.linalg.inv(K)
screen_pixel = h5data['screen_position'][i].reshape([-1,3])
target = screen_pixel
mask = h5data['mask'][i]
valid = screen_pixel[:,0] != 0
ray_origin, ray_dir = generate_ray(self.resy, self.resx, K_inverse, R_inverse)
mask = process_mask(mask)
target = torch.tensor(target , dtype = Float).pin_memory()
valid = torch.tensor(valid , dtype = bool ).pin_memory()
mask = torch.tensor(mask , dtype = Float).pin_memory()
ray_origin = torch.tensor(ray_origin , dtype = Float).pin_memory()
ray_dir = torch.tensor(ray_dir , dtype = Float).pin_memory()
R_inverse = torch.tensor(R_inverse , dtype = Float).pin_memory()
K_inverse = torch.tensor(K_inverse , dtype = Float).pin_memory()
R = torch.tensor(R , dtype = Float).pin_memory()
K = torch.tensor(K , dtype = Float).pin_memory()
camera_M = (R, K, R_inverse, K_inverse)
self.Views.append((target, valid, mask, ray_origin, ray_dir, camera_M))
h5data.close()