-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathrotate_bbox_one_by_one.py
228 lines (195 loc) · 11.8 KB
/
rotate_bbox_one_by_one.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
# 图像800x640的串行排列
# 任务是逐次旋转bbox对应区域
# 处理jpg 和 xml文件
# 按照源目录和标准命名备份
import xml.etree.ElementTree as ET
import os
import numpy as np
from os import getcwd
from PIL import Image
import cv2
import imgaug as ia
from imgaug import augmenters as iaa
from imgaug.augmentables.bbs import BoundingBox, BoundingBoxesOnImage
ia.seed(1)
def read_xml_annotation(image_id):
# members = tree.findall('object')
in_file = open(image_id)
tree = ET.parse(in_file)
root = tree.getroot()
box_list = []
objects = root.findall('object')
for object in objects:
bndbox = object.find('bndbox')
xmin = int(bndbox.find('xmin').text)
xmax = int(bndbox.find('xmax').text)
ymin = int(bndbox.find('ymin').text)
ymax = int(bndbox.find('ymax').text)
box_list.append((xmin, ymin, xmax, ymax))
return box_list
def change_xml_annotation(xml_path, new_boxes, i, image_size, store_path):
# i the sequence number of this box
in_file = open(xml_path)
tree = ET.parse(in_file)
xmlroot = tree.getroot()
xmlroot.find('size')[0].text = str(image_size[1]) # width
xmlroot.find('size')[1].text = str(image_size[0]) # height
objects = xmlroot.findall('object')
bndbox = objects[i].find('bndbox')
points = objects[i].find('shape')[0]
bndbox.find('xmin').text = str(new_boxes[0])
bndbox.find('ymin').text = str(new_boxes[1])
bndbox.find('xmax').text = str(new_boxes[2])
bndbox.find('ymax').text = str(new_boxes[3])
points[0].text = str(new_boxes[0])
points[1].text = str(new_boxes[1])
points[2].text = str(new_boxes[2])
points[3].text = str(new_boxes[3])
name_str = "_objrotate_" + str(i) + "_" + str(rotate_angle[0])
tree.write(store_path + xml_path.split('\\')[-1][:-4] + name_str + '.xml')
def main(rotate_angle=[10, 10], gamma=0.1):
path_list = []
path_list_copy = []
sourece_path = r'D:\iqiyi\DATA\organized2020-11-heri2ver'
store_path = r'D:\iqiyi\DATA\organized2020-11-heri2ver_bbox_rotated'
if not os.path.isdir(store_path):
os.mkdir(store_path)
for i in os.listdir(sourece_path):
temp_path = sourece_path + "\\" + i
path_list.append(temp_path) # 保存绝对地址
path_list_copy.append(store_path + "\\" + i)
if not os.path.isdir(store_path + "\\" + i):
os.mkdir(store_path + "\\" + i)
while len(path_list):
temp = path_list[-1]
temp_copy = path_list_copy[-1]
path_list.pop(-1)
path_list_copy.pop(-1)
file_list = os.listdir(temp)
for i in file_list:
if os.path.isdir(temp + "\\" + i):
path_list.append(temp + "\\" + i)
path_list_copy.append(temp_copy + "\\" + i)
os.mkdir(temp_copy + "\\" + i)
else:
if i[-4:] == r'.xml':
jpg_name = temp + "\\" + i[:-4] + r'.jpg'
xml_name = jpg_name[:-4] + r'.xml'
print(jpg_name)
normal_store = temp_copy + "\\"
img = cv2.imread(jpg_name)
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
# image.shape 800 640 3 转置也不行？ shape[0] ver, shape[1] heri, shape[3] channels
seq_rotate = iaa.Sequential([
# iaa.Flipud(0.5), # vertically flip 20% of all images
# iaa.Multiply((1.2, 1.5)), # change brightness, doesn't affect BBs
iaa.Affine(
# translate_px={"x": 10, "y": 10},
# scale=(0.8, 0.95),
rotate=rotate_angle
) # translate by 40/60px on x/y axis, and scale to 50-70%, affects BBs
])
bndbox = read_xml_annotation(xml_name)
# 此处 坐标是【x，y】，而图片索引是【y，x】且y指向不明？？？
for i in range(len(bndbox)):
center_x = (bndbox[i][0] + bndbox[i][2]) // 2
center_y = (bndbox[i][1] + bndbox[i][3]) // 2
# 有些框出现在图片之外或者大部分在图片之外，很少，没有意义，直接略过
if center_y < 0 or center_x < 0:
continue
box_width = bndbox[i][2] - bndbox[i][0] + 1
box_height = bndbox[i][3] - bndbox[i][1] + 1
hypotenuse = int(np.sqrt(box_width**2 + box_height**2)) + 1
area_width = hypotenuse + int(box_width * gamma)
area_height = hypotenuse + int(box_height * gamma)
area_p_x = int(center_x - area_width / 2.0)
area_p_y = int(center_y - area_height / 2.0)
# area picture make
# 先把框中内容提取出来然后周围padding到期望的范围，不然可能会超出图片范围
area_img = np.zeros((area_width, area_height), dtype=np.uint8)
area_img = cv2.cvtColor(area_img, cv2.COLOR_GRAY2BGR)
# area_img [x, y]
box_area = img[bndbox[i][1]:bndbox[i][3], bndbox[i][0]:bndbox[i][2], :]
# box_area [y, x] picture!!!!
box_area = box_area.transpose((1, 0, 2))
# box_area [x, y]
box_in_area_x = int((area_width - box_width) // 2)
box_in_area_y = int((area_height - box_height) // 2)
print(box_area.shape)
print(box_in_area_x, box_in_area_y)
print(area_img.shape)
area_img[box_in_area_x: box_in_area_x + box_area.shape[0], box_in_area_y: box_in_area_y + box_area.shape[1], :] = box_area
# area done
area_img = area_img.transpose((1, 0, 2))
bbs = ia.BoundingBoxesOnImage([
ia.BoundingBox(x1=bndbox[i][0] - area_p_x, y1=bndbox[i][1] - area_p_y, x2=bndbox[i][2] - area_p_x, y2=bndbox[i][3] - area_p_y)
], shape=area_img.shape)
# image_before = bbs.draw_on_image(area_img, size=1)
# Image.fromarray(image_before).save(
# store_path + "/" + jpg_name.split('\\')[-1][:-4] + 'before.jpg')
seq_det = seq_rotate.to_deterministic() # 保持坐标和图像同步改变，而不是随机
image_aug = seq_det.augment_images([area_img])[0]
bbs_aug = seq_det.augment_bounding_boxes([bbs])[0]
# for test
image_after = bbs_aug.draw_on_image(image_aug, size=1)
# Image.fromarray(image_after).save(store_path + "/" + jpg_name.split('\\')[-1][:-4] + 'after.jpg')
# make up the final picture using image_aug, img, area_p_x, area_p_y
# img [y, x] image_aug[x, y]
# image_after = image_after.transpose((1, 0, 2))
print(jpg_name)
aa = area_p_y + int(image_after.shape[0])
bb = area_p_x + int(image_after.shape[1])
if area_p_x >= 0 and area_p_y >= 0:
if aa <= img.shape[0] and bb <= img.shape[1]:
img[area_p_y:aa, area_p_x:bb, :] = image_after
elif aa <= img.shape[0] and bb > img.shape[1]:
img[area_p_y:aa, area_p_x:, :] = image_after[:, :img.shape[1] - area_p_x, :]
elif aa > img.shape[0] and bb <= img.shape[1]:
img[area_p_y:, area_p_x:bb, :] = image_after[:img.shape[0] - area_p_y, :, :]
else:
img[area_p_y:, area_p_x:, :] = image_after[:img.shape[0] - area_p_y, :img.shape[1] - area_p_x, :]
elif area_p_x < 0 and area_p_y >= 0:
if aa <= img.shape[0] and bb <= img.shape[1]:
img[area_p_y:aa, 0:bb, :] = image_after[:, -area_p_x:, :]
elif aa <= img.shape[0] and bb > img.shape[1]:
img[area_p_y:aa, 0:, :] = image_after[:, -area_p_x:img.shape[1] - area_p_x, :]
elif aa > img.shape[0] and bb <= img.shape[1]:
img[area_p_y:, 0:bb, :] = image_after[:img.shape[0] - area_p_y, -area_p_x:, :]
else:
img[area_p_y:, 0:, :] = image_after[:img.shape[0] - area_p_y, -area_p_x:img.shape[1] - area_p_x, :]
elif area_p_x >= 0 and area_p_y < 0:
if aa <= img.shape[0] and bb <= img.shape[1]:
img[0:aa, area_p_x:bb, :] = image_after[-area_p_y:, :, :]
elif aa <= img.shape[0] and bb > img.shape[1]:
print(area_p_y, aa)
img[0:aa, area_p_x:, :] = image_after[-area_p_y:, :img.shape[1] - area_p_x, :]
elif aa > img.shape[0] and bb <= img.shape[1]:
img[0:, area_p_x:bb, :] = image_after[-area_p_y:img.shape[0] - area_p_y, :, :]
else:
img[0:, area_p_x:, :] = image_after[-area_p_y:img.shape[0] - area_p_y, :img.shape[1] - area_p_x, :]
else:
if aa <= img.shape[0] and bb <= img.shape[1]:
img[0:aa, 0:bb, :] = image_after[-area_p_y:, -area_p_x:, :]
elif aa <= img.shape[0] and bb > img.shape[1]:
img[0:aa, 0:, :] = image_after[-area_p_y:, -area_p_x:img.shape[1] - area_p_x, :]
elif aa > img.shape[0] and bb <= img.shape[1]:
img[0:, 0:bb, :] = image_after[-area_p_y:img.shape[0] - area_p_y, -area_p_x:, :]
else:
img[0:, 0:, :] = image_after[-area_p_y:img.shape[0] - area_p_y, -area_p_x:img.shape[1] - area_p_x, :]
name_str = "_objrotate_" + str(i) + "_" + str(rotate_angle[0])
Image.fromarray(img).save(normal_store + "\\" + jpg_name.split('\\')[-1][:-4] + name_str + ".jpg")
# check and fix the bbox that locates outside the pix
# bbs_aug.remove_out_of_image().clip_out_of_image()
img_wh = img.shape[:2]
new_bndbox = []
print(img_wh[0], img_wh[1])
new_bndbox.append(max(int(bbs_aug.bounding_boxes[0].x1 + area_p_x), 0))
new_bndbox.append(max(int(bbs_aug.bounding_boxes[0].y1 + area_p_y), 0))
new_bndbox.append(min(int(bbs_aug.bounding_boxes[0].x2 + area_p_x), img_wh[1]))
new_bndbox.append(min(int(bbs_aug.bounding_boxes[0].y2 + area_p_y), img_wh[0]))
# 修改xml tree 并保存
change_xml_annotation(xml_name, new_bndbox, i, img_wh, normal_store)
# 超参数 ： 外扩区域比例 gamma
gamma = 0.1
rotate_angle = [10, 10]
main(rotate_angle, gamma)