-
Notifications
You must be signed in to change notification settings - Fork 0
/
DataPretreatment.py
211 lines (193 loc) · 11.5 KB
/
DataPretreatment.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
import os
import traceback
from PIL import Image
import numpy as np
import utils
image_path = r'F:\data\img_celeba'
# 原始数据
box_label_path = r'F:\data\list_bbox_celeba.txt'
# 边框标签
landmark_label = r'F:\data\list_landmarks_celeba.txt'
# 关键点标签
negative_path = r'F:\data\negative'
save_path = r'E:\data\data'
for image_size in [12, 24, 48]:
# 图片的尺寸是12，24，48的
positive_img_path = os.path.join(save_path, str(image_size), "positive")
part_img_path = os.path.join(save_path, str(image_size), "part")
negative_img_path = os.path.join(save_path, str(image_size), "negative")
landmark_img_path = os.path.join(save_path, str(image_size), "landmark")
for paths in [positive_img_path, part_img_path, negative_img_path, landmark_img_path]:
if not os.path.exists(paths):
os.makedirs(paths)
positive_label_path = os.path.join(save_path, str(image_size), "positive.txt")
part_label_path = os.path.join(save_path, str(image_size), "part.txt")
negative_label_path = os.path.join(save_path, str(image_size), "negative.txt")
landmark_label_path = os.path.join(save_path, str(image_size), "landmark.txt")
positive_count = 0
part_count = 0
negative_count = 1
landmark_count = 0
try:
positive_label_file = open(positive_label_path, "w")
part_label_file = open(part_label_path, "w")
negative_label_file = open(negative_label_path, "w")
landmark_label_file = open(landmark_label_path, "w")
a = -1
# 曾经的坑
# for j, line2 in enumerate(open(landmark_label)):
# for i, line1 in enumerate(open(box_label_path)):
with open(box_label_path) as line1s:
with open(landmark_label) as line2s:
for (line1, line2) in zip(line1s, line2s):
# 踩过的坑
# print(line1 + "line1")
# print(a)
# line1 = line1s.readlines()
# print(line1)
# line2 = line2s.readline()
# print(line2 + "line2")
# line1 = enumerate(line1s.readlines())
# line2 = enumerate(line2s.readlines())
a += 1
if a < 2:
continue
try:
strs1 = str(line1).strip().split()
# print(strs1)
# .strip()是去除头尾的多余空格
strs2 = str(line2).strip().split()
# # image_name_file = strs[0]
# strs = line.strip().split(",")
# strs = list(strs)
image_file_path = os.path.join(image_path, "{0}".format(strs1[0]))
# 根据标签文本里取到的图片名称找到路径，再找到图片和坐标点框出人脸
with Image.open(image_file_path) as img:
img_w, img_h = img.size
img = img.convert('RGB')
x1 = float(strs1[1])
y1 = float(strs1[2])
w = float(strs1[3])
h = float(strs1[4])
x2 = float(x1 + w)
y2 = float(y1 + h)
leye_x = float(strs2[1])
leye_y = float(strs2[2])
reye_x = float(strs2[3])
reye_y = float(strs2[4])
nose_x = float(strs2[5])
nose_y = float(strs2[6])
lmouth_x = float(strs2[7])
lmouth_y = float(strs2[8])
rmouth_x = float(strs2[9])
rmouth_y = float(strs2[10])
if max(w, h) < 40 or x1 < 0 or y1 < 0 or w < 0 or h < 0:
continue
boxes = [[x1, y1, x2, y2]]
cx = x1 + w / 2
cy = y1 + h / 2
# 计算人脸框中心点坐标
# 接下来就是创造部分样本的时刻了,不加倍了
for _ in range(1):
w_ = np.random.randint(-w * 0.2, w * 0.2)
h_ = np.random.randint(-h * 0.2, h * 0.2)
cx_ = cx + w_
cy_ = cy + h_
# 现在是让真实框的中心点坐标偏一下
# 下面，我们给他随机一个正方形边框
side_length = np.random.randint(int(0.8 * min(w, h)), np.ceil(1.25 * (max(w, h))))
# int 是向下取整，np.ceil 是向上取整。
# 边长取出来就可以计算坐标了
x1_ = np.max(cx_ - side_length / 2, 0)
# 坐标要是小于零就要跑到图片外面去了，这怎么可以呢
y1_ = np.max(cy_ - side_length / 2, 0)
x2_ = x1_ + side_length
y2_ = y1_ + side_length
# 存一下整理好的坐标吧
crop_box = np.array([x1_, y1_, x2_, y2_])
# 计算一下生成的坐标偏移量，这里跟老师的不一样，但自我感觉我的对啊
# 偏移量谁减谁都可以，只要反算的时候对应就好
# 左上角相对于左上角，右下角相对于右下角
offset_x1 = (x1_ - x1) / side_length
offset_y1 = (y1_ - y1) / side_length
offset_x2 = (x2_ - x2) / side_length
offset_y2 = (y2_ - y2) / side_length
# 关键点的偏移量都是相对于左上角的
ldmk_off1 = (leye_x - x1) / side_length
ldmk_off2 = (leye_y - y1) / side_length
ldmk_off3 = (reye_x - x1) / side_length
ldmk_off4 = (reye_y - y1) / side_length
ldmk_off5 = (nose_x - x1) / side_length
ldmk_off6 = (nose_y - y1) / side_length
ldmk_off7 = (lmouth_x - x1) / side_length
ldmk_off8 = (lmouth_y - y1) / side_length
ldmk_off9 = (rmouth_x - x1) / side_length
ldmk_off10 = (rmouth_y - y1) / side_length
# 位置找好了就可以切了
crop = img.crop(crop_box)
face_size = crop.resize((image_size, image_size))
# resize和reshape要用np.array型才能调用，list类型不能调用这个方法
# 做一下iou操作
ious = utils.ioufunction(crop_box, np.array(boxes))[0]
# 正样本
if ious > 0.65:
positive_label_file.write(
"positive/{0}.jpg {1} {2} {3} {4} {5} 0 0 0 0 0 0 0 0 0 0\n".format(positive_count, 1, offset_x1,
offset_y1, offset_x2, offset_y2))
# 写标签到文本中
positive_label_file.flush()
# 刷新文件缓冲池，让文件更新更快
# 将图片保存到对应文件里
face_size.save(os.path.join(positive_img_path, "{0}.jpg".format(positive_count)))
positive_count += 1
for _ in range(2): # 制作landmark
landmark_label_file.write(
"landmark/{0}.jpg {1} 0 0 0 0 {2} {3} {4} {5} {6} {7} {8} {9} {10} {11}\n".format(
landmark_count, 3, ldmk_off1, ldmk_off2, ldmk_off3, ldmk_off4, ldmk_off5,
ldmk_off6, ldmk_off7, ldmk_off8, ldmk_off9, ldmk_off10))
# 写标签到文本中
landmark_label_file.flush()
# 刷新文件缓冲池，让文件更新更快
# 将图片保存到对应文件里
face_size.save(os.path.join(landmark_img_path, "{0}.jpg".format(landmark_count)))
landmark_count += 1
# 部分样本
if ious > 0.4:
part_label_file.write(
"part/{0}.jpg {1} {2} {3} {4} {5} 0 0 0 0 0 0 0 0 0 0\n".format(part_count, 2, offset_x1, offset_y1,
offset_x2, offset_y2))
# 写标签到文本中
part_label_file.flush()
# 刷新文件缓冲池，让文件更新更快
# 将图片保存到对应文件里
face_size.save(os.path.join(part_img_path, "{0}.jpg".format(part_count)))
part_count += 1
except Exception as e:
traceback.print_exc()
# 制作负样本
negative_list = os.listdir(negative_path)
for i, nimg_name in enumerate(negative_list):
negative_path_ = os.path.join(negative_path, nimg_name)
with Image.open(negative_path_) as nimg:
nimg = nimg.convert('RGB')
nimg_w, nimg_h = nimg.size
if min(nimg_w, nimg_h) < 48:
# 这里会删除一部分图片
continue
for _ in range(150):
fx = np.random.randint(0.5, nimg_w * 1.)
fy = fx
crop_box = np.array([0, 0, fx, fy])
if min(nimg_w, nimg_h) < 48:
continue
crop = nimg.crop(crop_box)
face_size = crop.resize((image_size, image_size))
negative_label_file.write("negative/{0}.jpg 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0\n".format(negative_count))
negative_label_file.flush()
face_size.save(os.path.join(negative_img_path, "{0}.jpg".format(negative_count)))
negative_count += 1
finally:
positive_label_file.close()
negative_label_file.close()
part_label_file.close()
landmark_label_file.close()