-
Notifications
You must be signed in to change notification settings - Fork 1
/
generate_data.py
229 lines (208 loc) · 8.51 KB
/
generate_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
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
import tensorflow as tf
import numpy as np
import cv2
from config.config import config
import math
import os
# config = config.config
from scipy import misc
from tools import euler_angles_utils
def DateSet(file_list, args, debug=False, img_dir=config.drewimgdir_all_ssd, isTest=False):
mean_lmk = None
if args.mean_pts is not None:
mean_lmk = np.loadtxt(args.mean_pts)
file_list, landmarks, attributes, euler_angles, video_list, attr4 = gen_data(file_list, img_dir, args)
if debug:
n = args.batch_size * 10
file_list = file_list[:n]
landmarks = landmarks[:n]
attributes = attributes[:n]
euler_angles = euler_angles[:n]
dataset = tf.data.Dataset.from_tensor_slices((file_list, landmarks, attributes, euler_angles))
def random_rotate_image_func(image, lmk):
# img_draw1 = image.copy()
# img_draw1 = img_draw1.astype(np.uint8)
# img_draw1 = img_draw1[:,:,::-1]
# cv2.imwrite('/lp/random-transformed-img/000.jpg',img_draw1)
# np.savetxt('/lp/random-transformed-img/000.txt',lmk)
assert (image.shape == (config.imgsize, config.imgsize, args.image_channels))
# assert (lmk.shape == ())
# 旋转角度范围
angle = np.random.uniform(low=-30.0, high=30.0)
# angle = 20
rotated_img = misc.imrotate(image, angle, 'bicubic')
# print (rotated_img.dtype)
rotated_lmk = rotate_lmk_function(lmk, angle)
rotated_lmk = np.reshape(rotated_lmk, [config.point_size, 2])
key_lmk = rotated_lmk[config.mainPlaneLmkIndex]
pitch, yaw, roll = euler_angles_utils.calculate_pitch_yaw_roll(key_lmk, mean_lmk=mean_lmk)
# if img_count <= 10:
# img_draw = rotated_img.copy()
# img_draw = img_draw.astype(np.uint8)
# # np.copy()
# img_draw = img_draw[:, :, ::-1]
# # print(img_draw.shape)
# # for p in (rotated_lmk * config.imgsize).reshape([-1, 2]).astype(int):
# # cv2.circle(img_draw, tuple(p), 1, (0, 255, 0), -1)
# # cv2.circle(img_draw, tuple(p), 1, (0,255,0), -1)
# cv2.imwrite('/lp/random-transformed-img/001.jpg', img_draw)
# np.savetxt('/lp/random-transformed-img/001.txt',rotated_lmk)
# # exit()
# img_count += 1
return rotated_img, rotated_lmk.reshape([-1]), np.array(
[pitch * np.pi / 180, yaw * np.pi / 180, roll * np.pi / 180], dtype=np.float32)
def rotate_lmk_function(lmk, angle):
# print (lmk)
sin_angle = np.sin(-angle * np.pi / 180)
cos_angle = np.cos(-angle * np.pi / 180)
Emat = np.mat(np.eye(3, dtype=np.float32))
Rmat = np.mat(np.eye(3,dtype=np.float32))
Rmat[0, 0] = cos_angle
Rmat[1, 1] = cos_angle
Rmat[0, 1] = -sin_angle
Rmat[1, 0] = sin_angle
Tmat = Emat.copy()
Tmat[0:2, 2] = -0.5
mat = Tmat.I * Rmat * Tmat
lmk = np.reshape(lmk, [-1, 2])
lmkMat = np.ones([len(lmk), 3], dtype=np.float32)
lmkMat[:, 0:2] = lmk
lmkMat = np.mat(lmkMat)
lmkMat = lmkMat.T
lmk_rotated = mat * lmkMat
lmk_rotated = np.array(lmk_rotated.T)
rotated_lmk = lmk_rotated[:, 0:2].copy()
return rotated_lmk
def _parse_data(filename, landmarks, attributes, euler_angles):
# print (filename)
# filename, landmarks, attributes = data
file_contents = tf.read_file(filename)
image = tf.image.decode_png(file_contents, channels=args.image_channels)
# print (filename , image.shape)
# print(image.get_shape())
# image.set_shape((args.image_size, args.image_size, args.image_channels))
image = tf.image.resize_images(image, (args.image_size, args.image_size), method=0)
if not isTest and args.augument:
image, landmarks, euler_angles = tf.py_func(random_rotate_image_func, [image, landmarks],
[tf.uint8, tf.float32, tf.float32])
image = tf.image.random_brightness(image, 20.0/256)
# tf.image.random_brightness(image,
image = tf.cast(image, tf.float32)
image = image / 256.0
return image, landmarks, attributes, euler_angles
dataset = dataset.map(_parse_data)
if not isTest:
dataset = dataset.shuffle(buffer_size=10000)
return dataset, len(file_list)
def generateAttribute(landmark, attribute):
'''
:param landmark: 106 landmark
:param attribute: 4 attribute
:return: new attribute
'''
attr = np.zeros((10,), dtype=np.int32)
landmark = np.asarray(landmark, dtype=np.float32)
assert len(landmark) == 106 * 2, "len lmk %d " % (len(landmark))
distToLeft = np.sqrt(np.sum(np.square(landmark[config.noseTipIndex] - landmark[config.faceLeftIndex])))
distToRight = np.sqrt(np.sum(np.square(landmark[config.noseTipIndex] - landmark[config.faceRightIndex])))
if distToRight == 0.0:
distToRight = 0.001
if distToLeft == 0.0:
distToLeft = 0.001
# profile
if distToRight / distToLeft >= 4 or distToLeft / distToRight >= 4:
attr[0] = 1
else:
attr[1] = 1
# man or woman
if attribute[0] == 1:
attr[2] = 1
else:
attr[3] = 1
# open mouth or not
if attribute[2] == 1:
attr[4] = 1
else:
attr[5] = 1
# smile or not
if attribute[3] == 1:
attr[6] = 1
else:
attr[7] = 1
if attribute[1] == 1:
attr[8] = 1
else:
attr[9] = 1
return attr
def gen_data(file_list, img_dir=config.drewimgdir_all_ssd, args=None, video_paths_file=None):
print('reading file %s' % (file_list))
with open(file_list, 'r') as f:
lines = f.readlines()
filenames, landmarks, attributes10, euler_angles, attributes4 = [], [], [], [], []
for line in lines:
line = line.strip().split()
if not 224:
print('wrinf line num %d : %s ' % (len(line), line))
path = img_dir + '/' + line[0]
landmark = line[1:106 * 2 + 1]
attribute = line[106 * 2 + 1:106 * 2 + 1 + 4]
box = line[106 * 2 + 1 + 4:106 * 2 + 1 + 4 + 4]
euler_angle = line[106 * 2 + 1 + 4 + 4:]
# print (line)
try:
assert len(attribute) == 4, '%s' % (len(attribute))
except:
print(len(line))
input('???')
attribute = [attribute[0], attribute[1], attribute[2], attribute[3]]
try:
assert len(landmark) == 106 * 2
except:
print('line: %s' % line)
if args is None:
landmark = np.asarray(landmark, dtype=np.float32)
else:
if args.lmk_norm:
landmark = np.asarray(landmark, dtype=np.float32) / config.imgsize
# print (landmark)
# print (landmark)
else:
landmark = np.asarray(landmark, dtype=np.float32)
attribute = np.asarray(attribute, dtype=np.int32)
attributes4.append(attribute)
attribute10 = generateAttribute(landmark, attribute)
euler_angle = np.asarray(euler_angle, dtype=np.float32) * math.pi / 180
filenames.append(path)
landmarks.append(landmark)
attributes10.append(attribute10)
euler_angles.append(euler_angle)
video_paths = []
if video_paths_file is not None:
with open(video_paths_file, 'r') as video_file:
lines = video_file.readlines()
for line in lines:
video_paths.append(line.strip())
video_paths = np.array(video_paths, dtype=str)
if len(video_paths) == 0:
video_paths = None
filenames = np.asarray(filenames, dtype=np.str)
landmarks = np.asarray(landmarks, dtype=np.float32)
attributes10 = np.asarray(attributes10, dtype=np.int32)
attributes4 = np.asarray(attributes4, dtype=np.float32)
euler_angles = np.asarray(euler_angles, dtype=np.float32)
return filenames, landmarks, attributes10, euler_angles, video_paths, attributes4
if __name__ == '__main__':
file_list = 'data/train_data/list.txt'
filenames, landmarks, attributes = gen_data(file_list)
for i in range(len(filenames)):
filename = filenames[i]
landmark = landmarks[i]
attribute = attributes[i]
print(attribute)
img = cv2.imread(filename)
h, w, _ = img.shape
landmark = landmark.reshape(-1, 2) * [h, w]
for (x, y) in landmark.astype(np.int32):
cv2.circle(img, (x, y), 1, (0, 0, 255))
cv2.imshow('0', img)
cv2.waitKey(0)