/
data_handler.py
161 lines (121 loc) · 7.59 KB
/
data_handler.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
import numpy as np
import sys
import os
from scipy import misc
class Data_handler:
def __init__(self, data_version, data_root, util_root, num_tr_img, \
num_val_img, num_val_loc, batch_size, patch_size, disp_range):
if data_version == 'kitti2015':
self.num_channels = 3
elif data_version == 'kitti2012':
self.num_channels = 1
else:
sys.exit('data_version should be either kitti2012 or kitti2015')
self.batch_size = batch_size
self.patch_size = patch_size
self.disp_range = disp_range
self.half_patch = patch_size // 2
self.half_range = disp_range // 2
self.tr_ptr = 0
self.curr_epoch = 0
# Stored img id for validation dataset
self.img_id_arr = []
#
self.file_ids = np.fromfile(os.path.join(util_root, 'myPerm.bin'), '<f4')
self.tr_loc = np.fromfile(('%s/tr_%d_%d_%d.bin') %
(util_root, num_tr_img, self.half_patch, self.half_range), '<f4').reshape(-1, 5).astype(int)
if num_val_img == 0:
self.val_loc = self.tr_loc
print('validate on training set..')
else:
self.val_loc = np.fromfile(('%s/val_%d_%d_%d.bin') %
(util_root, num_val_img, self.half_patch, self.half_range), '<f4').reshape(-1,5).astype(int)
print(('#training locations: %d -- #valuation locations: %d') % (self.tr_loc.shape[0],
self.val_loc.shape[0]))
for i in range(2, 5):
self.tr_loc[:, i] -= 1
self.val_loc[:, i] -= 1
self.ldata = {}
self.rdata = {}
# SET SEED BEFORE CALLING THIS FUNCTION
self.tr_perm = np.arange(self.tr_loc.shape[0])
self.val_perm = np.arange(self.val_loc.shape[0])
np.random.shuffle(self.tr_perm)
np.random.shuffle(self.val_perm)
for i in range(num_tr_img+num_val_img):
fn = self.file_ids[i]
if data_version == 'kitti2015':
l_img = misc.imread(('%s/image_2/%06d_10.png') % (data_root, fn))
r_img = misc.imread(('%s/image_3/%06d_10.png') % (data_root, fn))
elif data_version == 'kitti2012':
l_img = misc.imread(('%s/image_0/%06d_10.png') % (data_root, fn))
r_img = misc.imread(('%s/image_1/%06d_10.png') % (data_root, fn))
l_img = (l_img - l_img.mean()) / l_img.std()
r_img = (r_img - r_img.mean()) / r_img.std()
self.ldata[fn] = l_img.reshape(l_img.shape[0], l_img.shape[1], self.num_channels)
self.rdata[fn] = r_img.reshape(r_img.shape[0], r_img.shape[1], self.num_channels)
self.batch_left = np.zeros((self.batch_size, self.patch_size, self.patch_size, self.num_channels))
self.batch_right = np.zeros((self.batch_size, self.patch_size, self.patch_size + self.disp_range - 1, self.num_channels))
# self.batch_label = np.full((self.batch_size, 1), self.half_range)
self.batch_label = np.zeros((self.batch_size, disp_range))
dist = [0.05, 0.2, 0.5, 0.2, 0.05]
half_dist = len(dist) // 2
count = 0
for i in range(disp_range //2 - half_dist, disp_range // 2 + half_dist + 1):
self.batch_label[:, i] = dist[count]
count += 1
self.val_left = np.zeros((num_val_loc, self.patch_size, self.patch_size, self.num_channels))
self.val_right = np.zeros((num_val_loc, self.patch_size, self.patch_size + self.disp_range - 1, self.num_channels))
# self.val_label = np.full((num_val_loc, 1), self.half_range)
self.val_label = np.zeros((num_val_loc, disp_range))
count = 0
for i in range(disp_range //2 - half_dist, disp_range // 2 + half_dist + 1):
self.val_label[:, i] = dist[count]
count += 1
for i in range(num_val_loc):
img_id, loc_type, center_x, center_y, right_center_x = self.val_loc[self.val_perm[i], 0], self.val_loc[self.val_perm[i], 1], self.val_loc[self.val_perm[i], 2], self.val_loc[self.val_perm[i], 3], self.val_loc[self.val_perm[i], 4]
right_center_y = center_y
# Add img_id to img_id_arr
self.img_id_arr.append(img_id)
# print(self.rdata[img_id].shape)
# print(self.val_loc[self.val_perm[i]])
self.val_left[i] = self.ldata[img_id][(center_y-self.half_patch) : (center_y+self.half_patch + 1), (center_x-self.half_patch) : (center_x+self.half_patch + 1), :]
if loc_type == 1: # horizontal
self.val_right[i] = self.rdata[img_id][right_center_y-self.half_patch : right_center_y+self.half_patch + 1, right_center_x-self.half_patch-self.half_range : right_center_x+self.half_patch+self.half_range + 1, :]
elif loc_type == 2: # vertical
self.val_right[i] = np.transpose(self.rdata[img_id][right_center_y-self.half_patch-self.half_range : right_center_y+self.half_patch+self.half_range + 1, right_center_x-self.half_patch : right_center_x+self.half_patch + 1, :], (1, 0, 2))
# Remove duplicate id
self.img_id_arr = np.unique(self.img_id_arr)
#print("ID: ", self.img_id_arr, "LEN ", len(self.img_id_arr))
# Save validation dataset to .npy file
with open('val.npy', 'wb') as f:
np.save(f, self.img_id_arr)
print('validation created: num(%d)' % num_val_loc)
def next_batch(self):
for idx in range(self.batch_size):
i = self.tr_ptr + (idx + 1)
if i > self.tr_perm.shape[0]:
i = 1
self.tr_ptr = -(idx + 1) + 1
self.curr_epoch = self.curr_epoch + 1
print('....epoch id: ' + self.curr_epoch + ' done ......\n')
i -= 1
img_id, loc_type, center_x, center_y, right_center_x = self.tr_loc[self.tr_perm[i]][0], self.tr_loc[self.tr_perm[i]][1], self.tr_loc[self.tr_perm[i]][2], self.tr_loc[self.tr_perm[i]][3], self.tr_loc[self.tr_perm[i]][4]
right_center_y = center_y
if loc_type == 1: # horizontal
self.batch_left[idx] = self.ldata[img_id][(center_y-self.half_patch) : (center_y+self.half_patch + 1), (center_x-self.half_patch) : (center_x+self.half_patch + 1), :]
self.batch_right[idx] = self.rdata[img_id][right_center_y-self.half_patch : right_center_y+self.half_patch + 1, right_center_x-self.half_patch-self.half_range : right_center_x+self.half_patch+self.half_range + 1, :]
elif loc_type == 2: # vertical
self.batch_left[idx] = np.transpose(self.ldata[img_id][(center_y-self.half_patch) : (center_y+self.half_patch + 1), (center_x-self.half_patch) : (center_x+self.half_patch + 1), :], (1, 0, 2))
self.batch_right[idx] = np.transpose(self.rdata[img_id][right_center_y-self.half_patch-self.half_range : right_center_y+self.half_patch+self.half_range + 1, right_center_x-self.half_patch : right_center_x+self.half_patch + 1, :], (1, 0, 2))
self.tr_ptr = self.tr_ptr + self.batch_size
return self.batch_left, self.batch_right, self.batch_label
def evaluate(self):
return self.val_left, self.val_right, self.val_label
if __name__ == '__main__':
dh = Data_handler(data_version='kitti2012', data_root=os.path.expanduser('~/scratch/algolux/training'),
util_root=os.path.expanduser('~/scratch/algolux/cvpr16_stereo_public/preprocess/debug_12'), num_tr_img=160,
num_val_img=34, num_val_loc=100, batch_size=128, patch_size=37, disp_range=201)
bleft, bright, blabels = dh.next_batch()
print(bleft.shape)
print(blabels[:5])