forked from Oreobird/tf-keras-deep-head-pose
-
Notifications
You must be signed in to change notification settings - Fork 0
/
datasets.py
352 lines (290 loc) · 13.2 KB
/
datasets.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
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
import os
import numpy as np
import cv2
import scipy.io as sio
import utils
import random
def split_samples(samples_file, train_file, test_file, ratio=0.8):
with open(samples_file) as samples_fp:
lines = samples_fp.readlines()
random.shuffle(lines)
train_num = int(len(lines) * ratio)
test_num = len(lines) - train_num
count = 0
data = []
for line in lines:
count += 1
data.append(line)
if count == train_num:
with open(train_file, "w+") as train_fp:
for d in data:
train_fp.write(d)
data = []
if count == train_num + test_num:
with open(test_file, "w+") as test_fp:
for d in data:
test_fp.write(d)
data = []
return train_num, test_num
def get_list_from_filenames(file_path):
with open(file_path) as f:
lines = f.read().splitlines()
return lines
class Biwi:
def __init__(self, data_dir, data_file, batch_size=64, input_size=64, ratio=0.8):
self.data_dir = data_dir
self.data_file = data_file
self.batch_size = batch_size
self.input_size = input_size
self.train_file = None
self.test_file = None
self.__gen_filename_list(os.path.join(self.data_dir, self.data_file))
self.train_num, self.test_num = self.__gen_train_test_file(os.path.join(self.data_dir, 'train.txt'),
os.path.join(self.data_dir, 'test.txt'), ratio=ratio)
def __get_input_img(self, data_dir, file_name, img_ext='.png', annot_ext='.txt'):
img = cv2.imread(os.path.join(data_dir, file_name + '_rgb' + img_ext))
bbox_path = os.path.join(data_dir, file_name.split('/')[0] + '/bbox.txt')
# Load bounding box
bbox = open(bbox_path, 'r')
line = bbox.readline().split(' ')
if len(line) < 4:
x_min, x_max, y_min, y_max = 0, img.size[0], 0, img.size[1]
else:
x_min, x_max, y_min, y_max = [float(line[0]), float(line[1]), float(line[2]), float(line[3])]
bbox.close()
# Loosely crop face
k = 0.3
x_min -= k * abs(x_max - x_min)
y_min -= k * abs(y_max - y_min)
x_max += k * abs(x_max - x_min)
y_max += k * abs(y_max - y_min)
crop_img = img[int(y_min): int(y_max), int(x_min): int(x_max)]
# print(crop_img.shape)
# cv2.imshow('crop_img', crop_img)
# cv2.waitKey(0)
crop_img = cv2.resize(crop_img, (self.input_size, self.input_size))
crop_img = np.asarray(crop_img)
normed_img = (crop_img - crop_img.mean())/crop_img.std()
return normed_img
def __get_input_label(self, data_dir, file_name, annot_ext='.txt'):
# Load pose in degrees
pose_path = os.path.join(data_dir, file_name + '_pose' + annot_ext)
pose_annot = open(pose_path, 'r')
R = []
for line in pose_annot:
line = line.strip('\n').split(' ')
l = []
if line[0] != '':
for nb in line:
if nb == '':
continue
l.append(float(nb))
R.append(l)
R = np.array(R)
T = R[3, :]
R = R[:3, :]
pose_annot.close()
R = np.transpose(R)
roll = -np.arctan2(R[1][0], R[0][0]) * 180 / np.pi
yaw = -np.arctan2(-R[2][0], np.sqrt(R[2][1] ** 2 + R[2][2] ** 2)) * 180 / np.pi
pitch = np.arctan2(R[2][1], R[2][2]) * 180 / np.pi
# Bin values
bins = np.array(range(-99, 99, 3))
binned_labels = np.digitize([yaw, pitch, roll], bins) - 1
cont_labels = [yaw, pitch, roll]
return binned_labels, cont_labels
def __gen_filename_list(self, filename_list_file):
if not os.path.exists(filename_list_file):
with open(filename_list_file, 'w+') as tlf:
for root, dirs, files in os.walk(self.data_dir):
for subdir in dirs:
subfiles = os.listdir(os.path.join(self.data_dir, subdir))
for f in subfiles:
if os.path.splitext(f)[1] == '.png':
token = os.path.splitext(f)[0].split('_')
filename = token[0] + '_' + token[1]
# print(filename)
tlf.write(subdir + '/' + filename + '\n')
def __gen_train_test_file(self, train_file, test_file, ratio=0.8):
self.train_file = train_file
self.test_file = test_file
return split_samples(os.path.join(self.data_dir, self.data_file), self.train_file, self.test_file, ratio=ratio)
def train_num(self):
return self.train_num
def test_num(self):
return self.test_num
def save_test(self, name, save_dir, prediction):
img_path = os.path.join(self.data_dir, name + '_rgb.png')
# print(img_path)
cv2_img = cv2.imread(img_path)
cv2_img = utils.draw_axis(cv2_img, prediction[0], prediction[1], prediction[2], tdx=200, tdy=200,
size=100)
save_path = os.path.join(save_dir, name.split('/')[1] + '.png')
# print(save_path)
cv2.imwrite(save_path, cv2_img)
def data_generator(self, shuffle=True, test=False):
sample_file = self.train_file
if test:
sample_file = self.test_file
filenames = get_list_from_filenames(sample_file)
file_num = len(filenames)
while True:
if shuffle and not test:
idx = np.random.permutation(range(file_num))
filenames = np.array(filenames)[idx]
max_num = file_num - (file_num % self.batch_size)
for i in range(0, max_num, self.batch_size):
batch_x = []
batch_yaw = []
batch_pitch = []
batch_roll = []
names = []
for j in range(self.batch_size):
img = self.__get_input_img(self.data_dir, filenames[i + j])
bin_labels, cont_labels = self.__get_input_label(self.data_dir, filenames[i + j])
#print(img.shape)
batch_x.append(img)
batch_yaw.append([bin_labels[0], cont_labels[0]])
batch_pitch.append([bin_labels[1], cont_labels[1]])
batch_roll.append([bin_labels[2], cont_labels[2]])
names.append(filenames[i + j])
batch_x = np.array(batch_x, dtype=np.float32)
batch_yaw = np.array(batch_yaw)
batch_pitch = np.array(batch_pitch)
batch_roll = np.array(batch_roll)
if test:
yield (batch_x, [batch_yaw, batch_pitch, batch_roll], names)
else:
yield (batch_x, [batch_yaw, batch_pitch, batch_roll])
if test:
break
class AFLW2000:
def __init__(self, data_dir, data_file, batch_size=16, input_size=64):
self.data_dir = data_dir
self.data_file = data_file
self.batch_size = batch_size
self.input_size = input_size
self.train_file = None
self.test_file = None
self.__gen_filename_list(os.path.join(self.data_dir, self.data_file))
self.train_num, self.test_num = self.__gen_train_test_file(os.path.join(self.data_dir, 'train.txt'),
os.path.join(self.data_dir, 'test.txt'))
def __get_ypr_from_mat(self, mat_path):
mat = sio.loadmat(mat_path)
pre_pose_params = mat['Pose_Para'][0]
pose_params = pre_pose_params[:3]
return pose_params
def __get_pt2d_from_mat(self, mat_path):
mat = sio.loadmat(mat_path)
pt2d = mat['pt2d']
return pt2d
def __get_input_img(self, data_dir, file_name, img_ext='.jpg', annot_ext='.mat'):
img = cv2.imread(os.path.join(data_dir, file_name + img_ext))
pt2d = self.__get_pt2d_from_mat(os.path.join(data_dir, file_name + annot_ext))
# Crop the face loosely
x_min = min(pt2d[0, :])
y_min = min(pt2d[1, :])
x_max = max(pt2d[0, :])
y_max = max(pt2d[1, :])
Lx = abs(x_max - x_min)
Ly = abs(y_max - y_min)
Lmax = max(Lx, Ly) * 1.5
center_x = x_min + Lx // 2
center_y = y_min + Ly // 2
x_min = center_x - Lmax // 2
x_max = center_x + Lmax // 2
y_min = center_y - Lmax // 2
y_max = center_y + Lmax // 2
if x_min < 0:
y_max -= abs(x_min)
x_min = 0
if y_min < 0:
x_max -= abs(y_min)
y_min = 0
if x_max > img.shape[1]:
y_min += abs(x_max - img.shape[1])
x_max = img.shape[1]
if y_max > img.shape[0]:
x_min += abs(y_max - img.shape[0])
y_max = img.shape[0]
# print("x_min:{},x_max:{},y_min:{},y_max{}".format(x_min, x_max, y_min, y_max))
crop_img = img[int(y_min):int(y_max), int(x_min):int(x_max)]
# print(crop_img.shape)
# cv2.imshow('crop_img', crop_img)
# cv2.waitKey(0)
crop_img = np.asarray(cv2.resize(crop_img, (self.input_size, self.input_size)))
normed_img = (crop_img - crop_img.mean()) / crop_img.std()
# print(normed_img)
return normed_img
def __get_input_label(self, data_dir, file_name, annot_ext='.mat'):
# We get the pose in radians
pose = self.__get_ypr_from_mat(os.path.join(data_dir, file_name + annot_ext))
# And convert to degrees.
yaw = pose[1] * 180.0 / np.pi
pitch = pose[0] * 180.0 / np.pi
roll = pose[2] * 180.0 / np.pi
cont_labels = [yaw, pitch, roll]
# print(cont_labels)
# Bin values
bins = np.array(range(-99, 99, 3))
bin_labels = np.digitize([yaw, pitch, roll], bins) - 1
return bin_labels, cont_labels
def __gen_filename_list(self, filename_list_file):
if not os.path.exists(filename_list_file):
with open(filename_list_file, 'w+') as tlf:
for root, dirs, files in os.walk(self.data_dir):
for f in files:
if os.path.splitext(f)[1] == '.jpg':
tlf.write(os.path.splitext(f)[0] + '\n')
def __gen_train_test_file(self, train_file, test_file):
self.train_file = train_file
self.test_file = test_file
return split_samples(os.path.join(self.data_dir, self.data_file), self.train_file, self.test_file, ratio=0.8)
def train_num(self):
return self.train_num
def test_num(self):
return self.test_num
def save_test(self, name, save_dir, prediction):
img_path = os.path.join(self.data_dir, name + '.jpg')
# print(img_path)
cv2_img = cv2.imread(img_path)
cv2_img = utils.draw_axis(cv2_img, prediction[0], prediction[1], prediction[2], tdx=200, tdy=200,
size=100)
save_path = os.path.join(save_dir, name + '.jpg')
# print(save_path)
cv2.imwrite(save_path, cv2_img)
def data_generator(self, shuffle=True, test=False):
sample_file = self.train_file
if test:
sample_file = self.test_file
filenames = get_list_from_filenames(sample_file)
file_num = len(filenames)
print(file_num)
while True:
if shuffle:
idx = np.random.permutation(range(file_num))
filenames = np.array(filenames)[idx]
max_num = file_num - (file_num % self.batch_size)
for i in range(0, max_num, self.batch_size):
batch_x = []
batch_yaw = []
batch_pitch = []
batch_roll = []
names = []
for j in range(self.batch_size):
img = self.__get_input_img(self.data_dir, filenames[i + j])
bin_labels, cont_labels = self.__get_input_label(self.data_dir, filenames[i + j])
# print(img.shape)
batch_x.append(img)
batch_yaw.append([bin_labels[0], cont_labels[0]])
batch_pitch.append([bin_labels[1], cont_labels[1]])
batch_roll.append([bin_labels[2], cont_labels[2]])
names.append(filenames[i + j])
batch_x = np.array(batch_x, dtype=np.float32)
batch_yaw = np.array(batch_yaw)
batch_pitch = np.array(batch_pitch)
batch_roll = np.array(batch_roll)
if test:
yield (batch_x, [batch_yaw, batch_pitch, batch_roll], names)
else:
yield (batch_x, [batch_yaw, batch_pitch, batch_roll])