-
Notifications
You must be signed in to change notification settings - Fork 0
/
generate_data_coco.py
512 lines (419 loc) · 21.9 KB
/
generate_data_coco.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
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
import os
import cv2
import numpy.random as random
import numpy as np
import copy
from PIL import Image, ImageDraw
import xml.etree.ElementTree as ET
import json
""" generate_dataset = generate_dataset(dir_list列表[‘label-1靶标图片相对于项目绝对地址的相对地址’,...], label_list列表['label-1',...],
'DIV2K_valid_HR'背景图片存储地址, r'D:\pythonproject\generatedataset'项目的绝对地址,
'DIV2K_Grey/valid'图片保存地址, 'DIV2K_Grey/trash'xml文件保存地址)
从dir_list随机选取相对路径路径与项目绝对地址组合读取png靶标,根据索引对应label_list标签.
,从background文件夹读取png或jpg背景,背景均匀划分为四块,将四张target图像分别覆盖在四个区域.
生成图片保存在生成'DIV2K_Grey/valid'
voc格式xml文件保存在'DIV2K_Grey/trash'
"""
def annotate_image(image, keypoints,label):
color_dis = {'circle':(0, 255, 0), 'cross':(255, 0, 0), 't':(0, 0, 255)}
color = color_dis[label]
for keypoint in keypoints:
x, y = keypoint
cv2.circle(image, (int(x), int(y)), 3, color, -1)
return image
def calculate_iou(lis1, lis2):
# 计算相交区域的左上角和右下角坐标
xmin1, ymin1, xmax1, ymax1 = lis1
xmin2, ymin2, xmax2, ymax2 = lis2
inter_xmin = max(xmin1, xmin2)
inter_ymin = max(ymin1, ymin2)
inter_xmax = min(xmax1, xmax2)
inter_ymax = min(ymax1, ymax2)
# 计算相交区域的宽度和高度
inter_width = inter_xmax - inter_xmin
inter_height = inter_ymax - inter_ymin
# 如果相交区域的宽度或高度小于等于0,表示两个矩形框没有重叠,返回0作为IoU
if inter_width <= 0 or inter_height <= 0:
return 0.0
# 计算相交区域的面积
inter_area = inter_width * inter_height
# 计算两个矩形框的面积
area1 = (xmax1 - xmin1) * (ymax1 - ymin1)
area2 = (xmax2 - xmin2) * (ymax2 - ymin2)
# 计算并集的面积
union_area = area1 + area2 - inter_area
# 计算IoU
iou = inter_area / union_area
return iou
class generate_dataset(object):
def __init__(self, target_dir_path_list, label_list, background_dir_path,
project_abspath, image_save_path, annotation_save_path, json_file_dir):
self.target_dir_path_list = target_dir_path_list
self.label_list = label_list
self.background_dir_path = background_dir_path
self.project_abspath = project_abspath
self.image_save_path = image_save_path
self.annotation_save_path = annotation_save_path
self.classes = len(self.label_list)
self.target_size_range = [0.1, 0.25] # 靶标大小的区间,占背景图最短边的比例[0.06, 0.25]
self.target_filename_list = []
self.background_filename_list = [os.path.join(self.background_dir_path, f) for f in
os.listdir(self.background_dir_path) if
f.endswith('.png') or f.endswith('.jpg')]
self.json_file_dir = json_file_dir
self.seize_filename()
self.i = 0
def seize_filename(self):
# 获取filrname list 【【】,【】,【】】
for target_dir in self.target_dir_path_list:
self.target_filename_list.append([os.path.join(target_dir, f) for f in os.listdir(target_dir) if
f.endswith('.png')])
def add_alpha_channel(self, img):
""" 为jpg图像添加alpha通道 """
b_channel, g_channel, r_channel = cv2.split(img) # 剥离jpg图像通道
alpha_channel = np.ones(b_channel.shape, dtype=b_channel.dtype) * 255 # 创建Alpha通道
img_new = cv2.merge((b_channel, g_channel, r_channel, alpha_channel)) # 融合通道
return img_new
def merge_img(self, jpg_img, png_img, y1, y2, x1, x2):
""" 将png透明图像与jpg图像叠加
y1,y2,x1,x2为叠加位置坐标值
"""
# cv2.imshow(png_img)
# cv2.imshow(jpg_img)
# 判断jpg图像是否已经为4通道
if jpg_img.shape[2] == 3:
jpg_img = self.add_alpha_channel(jpg_img)
'''
当叠加图像时,可能因为叠加位置设置不当,导致png图像的边界超过背景jpg图像,而程序报错
这里设定一系列叠加位置的限制,可以满足png图像超出jpg图像范围时,依然可以正常叠加
'''
yy1 = 0
yy2 = png_img.shape[0]
xx1 = 0
xx2 = png_img.shape[1]
if x1 < 0:
xx1 = -x1
x1 = 0
if y1 < 0:
yy1 = - y1
y1 = 0
if x2 > jpg_img.shape[1]:
xx2 = png_img.shape[1] - (x2 - jpg_img.shape[1])
x2 = jpg_img.shape[1]
if y2 > jpg_img.shape[0]:
yy2 = png_img.shape[0] - (y2 - jpg_img.shape[0])
y2 = jpg_img.shape[0]
# 获取要覆盖图像的alpha值,将像素值除以255,使值保持在0-1之间
# if not (png_img.shape[2] == 4):
# print(png_img.shape)
# print('-------------')
# cv2.imshow('qq', png_img)
# cv2.waitKey(0)
alpha_png = png_img[yy1:yy2, xx1:xx2, 3] / 255.0
alpha_jpg = 1 - alpha_png
# 开始叠加
for c in range(0, 3):
jpg_img[y1:y2, x1:x2, c] = (
(alpha_jpg * jpg_img[y1:y2, x1:x2, c]) + (alpha_png * png_img[yy1:yy2, xx1:xx2, c]))
return jpg_img
def subarea(self, bg_h, bg_w, target_size):
""" 划分四个区域
"""
target_gap = target_size
# assert(target_size<bg_h//2 and target_size<bg_w//2)
subarea_dict = {'sub1': [0, bg_w // 2 - target_gap, 0, bg_h // 2 - target_gap],
'sub2': [bg_w // 2, bg_w - target_gap, 0, bg_h // 2 - target_gap],
'sub3': [0, bg_w // 2 - target_gap, bg_h // 2, bg_h - target_gap],
'sub4': [bg_w // 2, bg_w - target_gap, bg_h // 2, bg_h - target_gap]}
# [xmin,xmax,ymin,ymax]
return subarea_dict
def imread_image(self):
""" 读取图片
"""
choise_list = []
label_list = np.array(self.label_list)
num_images = 5
choise_list = random.randint(0, self.classes, size=num_images, dtype=int)
random_images = random.choice(self.target_filename_list[choise_list[0]])
self.img1 = cv2.imread(random_images, cv2.IMREAD_UNCHANGED)
filename1 = os.path.splitext(os.path.basename(random_images))[0]
json_path1 =os.path.join(self.json_file_dir, filename1 + '.json')
self.shapes1 = self.read_json(json_path1)
random_images = random.choice(self.target_filename_list[choise_list[1]])
self.img2 = cv2.imread(random_images, cv2.IMREAD_UNCHANGED)
filename2 = os.path.splitext(os.path.basename(random_images))[0]
json_path2 = os.path.join(self.json_file_dir, filename2 + '.json')
self.shapes2 = self.read_json(json_path2)
random_images = random.choice(self.target_filename_list[choise_list[2]])
self.img3 = cv2.imread(random_images, cv2.IMREAD_UNCHANGED)
filename3 = os.path.splitext(os.path.basename(random_images))[0]
json_path3 = os.path.join(self.json_file_dir, filename3 + '.json')
self.shapes3 = self.read_json(json_path3)
random_images = random.choice(self.target_filename_list[choise_list[3]])
self.img4 = cv2.imread(random_images, cv2.IMREAD_UNCHANGED)
filename4 = os.path.splitext(os.path.basename(random_images))[0]
json_path4 = os.path.join(self.json_file_dir, filename4 + '.json')
self.shapes4 = self.read_json(json_path4)
random_images = random.choice(self.target_filename_list[choise_list[4]])
self.img5 = cv2.imread(random_images, cv2.IMREAD_UNCHANGED)
filename5 = os.path.splitext(os.path.basename(random_images))[0]
json_path5 = os.path.join(self.json_file_dir, filename5 + '.json')
self.shapes5 = self.read_json(json_path5)
self.background = cv2.imread(self.background_filename_list[self.i])
self.img_label = label_list[choise_list]
self.img_label = self.img_label.tolist()
# 使用os.path.basename()获取文件名(包含后缀)
filename_with_extension = os.path.basename(self.background_filename_list[self.i])
# 使用os.path.splitext()获取文件名和后缀的分割结果
filename = os.path.splitext(filename_with_extension)[0]
return filename # 返回当前读取背景的图片名,
# cv2.imshow('a', self.background)
# cv2.waitKey(0)
def random_resize(self):
""" target图片随机变换大小返回最长边
"""
maxlength = 0
bg_minlength = min(self.background.shape[0], self.background.shape[1])
target_size_range = [0, 0]
target_size_range[0] = int(self.target_size_range[0] * bg_minlength)
target_size_range[1] = int(self.target_size_range[1] * bg_minlength)
h, w = self.img1.shape[0:2]
h_a = random.randint(target_size_range[0], target_size_range[1])
factor = h_a / h
w_a = int(factor * w)
maxlength = max(h_a, w_a, maxlength)
self.img1 = cv2.resize(self.img1, dsize=(w_a, h_a))
self.shapes1[0]["points"] = (np.array(self.shapes1[0]["points"]) * factor).tolist()
self.shapes1[1]["points"] = (np.array(self.shapes1[1]["points"]) * factor).tolist()
h, w = self.img2.shape[0:2]
h_a = random.randint(target_size_range[0], target_size_range[1])
factor = h_a / h
w_a = int(factor * w)
maxlength = max(h_a, w_a, maxlength)
self.img2 = cv2.resize(self.img2, dsize=(w_a, h_a))
self.shapes2[0]["points"] = (np.array(self.shapes2[0]["points"]) * factor).tolist()
self.shapes2[1]["points"] = (np.array(self.shapes2[1]["points"]) * factor).tolist()
h, w = self.img3.shape[0:2]
h_a = random.randint(target_size_range[0], target_size_range[1])
factor = h_a / h
w_a = int(factor * w)
maxlength = max(h_a, w_a, maxlength)
self.img3 = cv2.resize(self.img3, dsize=(w_a, h_a))
self.shapes3[0]["points"] = (np.array(self.shapes3[0]["points"]) * factor).tolist()
self.shapes3[1]["points"] = (np.array(self.shapes3[1]["points"]) * factor).tolist()
h, w = self.img4.shape[0:2]
h_a = random.randint(target_size_range[0], target_size_range[1])
factor = h_a / h
w_a = int(factor * w)
maxlength = max(h_a, w_a, maxlength)
self.img4 = cv2.resize(self.img4, dsize=(w_a, h_a))
self.shapes4[0]["points"] = (np.array(self.shapes4[0]["points"]) * factor).tolist()
self.shapes4[1]["points"] = (np.array(self.shapes4[1]["points"]) * factor).tolist()
h, w = self.img5.shape[0:2]
h_a = random.randint(target_size_range[0], target_size_range[1])
factor = h_a / h
w_a = int(factor * w)
maxlength = max(h_a, w_a, maxlength)
self.img5 = cv2.resize(self.img5, dsize=(w_a, h_a))
self.shapes5[0]["points"] = (np.array(self.shapes5[0]["points"]) * factor).tolist()
self.shapes5[1]["points"] = (np.array(self.shapes5[1]["points"]) * factor).tolist()
return maxlength
def voc_annotation(self, objname_list, filename, location_list):
""" 根据target插入位置编写voc格式xml文件
"""
height, width, channel = self.background.shape
xml_root = ET.Element("annotation")
ET.SubElement(xml_root, "folder").text = self.image_save_path
ET.SubElement(xml_root, "filename").text = filename
ET.SubElement(xml_root, "path").text = os.path.join(self.project_abspath, self.image_save_path,
filename + '.jpg')
source_node = ET.SubElement(xml_root, "source")
ET.SubElement(source_node, "database").text = "Unknown"
size_node = ET.SubElement(xml_root, "size")
ET.SubElement(size_node, "width").text = str(width)
ET.SubElement(size_node, "height").text = str(height)
ET.SubElement(size_node, "depth").text = str(channel)
self.test = copy.copy(self.background)
for index, location in enumerate(location_list):
x_min, y_min, x_max, y_max = location
cv2.rectangle(self.test, (x_min, y_min), (x_max, y_max), (0, 0, 255), 1)
object_node = ET.SubElement(xml_root, "object")
ET.SubElement(object_node, "name").text = objname_list[index]
bndbox_node = ET.SubElement(object_node, "bndbox") # 添加边界框节点
ET.SubElement(bndbox_node, "xmin").text = str(x_min)
ET.SubElement(bndbox_node, "ymin").text = str(y_min)
ET.SubElement(bndbox_node, "xmax").text = str(x_max)
ET.SubElement(bndbox_node, "ymax").text = str(y_max)
xml_file_path = os.path.join(self.annotation_save_path, filename + '.xml')
tree = ET.ElementTree(xml_root)
tree.write(xml_file_path, encoding="utf-8")
def random_set_img(self,location_list,overlap_rate):
x1, y1, x2, y2 = location_list[random.randint(len(location_list))]
w1 = x2 - x1
h1 = y2 - y1
h2, w2 = self.img5.shape[:2]
x_in_1 = x1 + overlap_rate * w1 - w2
y_in_1 = y1 + overlap_rate * h1 - h2
x_in_2 = x2 - overlap_rate * w1
y_in_2 = y2 - overlap_rate * h1
x_ex_1 = x1 - overlap_rate * w1 - w2
y_ex_1 = y1 - overlap_rate * h1 - h2
x_ex_2 = x2 + overlap_rate * w1
y_ex_2 = y2 + overlap_rate * h1
bg_h, bg_w = self.background.shape[0], self.background.shape[1]
while True:
x = random.randint(x_ex_1, x_ex_2)
y = random.randint(y_ex_1, y_ex_2)
if x < 0 or x > bg_w - w2 or y < 0 or y > bg_h - h2:
continue
if x_in_1 < x < x_in_2 and y_in_1 < y < y_in_2:
continue
return x, y, x+w2, y+h2
def read_json(self,json_path):
"return : dict with key-lable,points"
with open(json_path, 'r') as file:
data = json.load(file)
shapes = data["shapes"]
return shapes
def coco_annotation(self, filename, location_list):
img_path = os.path.join(self.project_abspath, self.image_save_path,
filename + '.jpg')
info_dict = {"version": "5.2.1",
"flags": {},
"shapes": [],
"imagePath": img_path,
"imageHeight": self.background.shape[0],
"imageWidth": self.background.shape[1]}
shape_list = [self.shapes1, self.shapes2, self.shapes3, self.shapes4, self.shapes5]
assert len(location_list) == len(shape_list) == 5
# 起始点位置决定shapes位置
for index, location in enumerate(location_list):
x_min, y_min, x_max, y_max = location
shapes = shape_list[index]
shapes[0]["points"] = (np.array(shapes[0]["points"]) + np.array([x_min, y_min])).tolist()
shapes[1]["points"] = (np.array(shapes[1]["points"]) + np.array([x_min, y_min])).tolist()
info_dict["shapes"].append(shapes[0])
info_dict["shapes"].append(shapes[1])
json_file_path = os.path.join(self.annotation_save_path, filename + '.json')
self.test = copy.copy(self.background)
for item in info_dict["shapes"]:
label = item["label"]
points = item["points"]
self.test = annotate_image(self.test, points, label)
# cv2.imshow('test', self.test)
# cv2.waitKey(0)
with open(json_file_path, 'w', encoding='utf-8') as file:
json.dump(info_dict, file, indent=2)
file.close()
def main(self):
while self.i < len(self.background_filename_list):
location_list = []
background_prefix = self.imread_image() # 读取target和背景图,返回当前背景图名字前缀
while not (self.img1.shape[2] == self.img2.shape[2] == self.img3.shape[2] == self.img4.shape[2] == self.img5.shape[2] == 4):
background_prefix = self.imread_image()
if not isinstance(self.background, np.ndarray):
self.i += 1
continue
if self.background.shape[0] < 500 and self.background.shape[1] < 500:
self.i += 1
continue
gap = self.random_resize()
subarea_dict = self.subarea(self.background.shape[0], self.background.shape[1], gap)
# print(subarea_dict)
print(self.background.shape)
# sub1 img1
x_min, x_max, y_min, y_max = subarea_dict['sub1']
x1 = random.randint(x_min, x_max)
y1 = random.randint(y_min, y_max)
x2 = x1 + self.img1.shape[1]
y2 = y1 + self.img1.shape[0]
self.background = self.merge_img(self.background, self.img1, y1, y2, x1, x2)
self.background = self.background[:, :, 0:3]
location_list.append([x1, y1, x2, y2])
# print('img1', self.img1.shape)
# print("x1,y1,x2,y2: ", x1, y1, x2, y2)
# sub2 img2
x_min, x_max, y_min, y_max = subarea_dict['sub2']
x1 = random.randint(x_min, x_max)
y1 = random.randint(y_min, y_max)
x2 = x1 + self.img2.shape[1]
y2 = y1 + self.img2.shape[0]
self.background = self.merge_img(self.background, self.img2, y1, y2, x1, x2)
self.background = self.background[:, :, 0:3]
location_list.append([x1, y1, x2, y2])
# print('img2',self.img2.shape)
# print("x1,y1,x2,y2: ", x1, y1, x2, y2)
# sub3 img3
x_min, x_max, y_min, y_max = subarea_dict['sub3']
x1 = random.randint(x_min, x_max)
y1 = random.randint(y_min, y_max)
x2 = x1 + self.img3.shape[1]
y2 = y1 + self.img3.shape[0]
self.background = self.merge_img(self.background, self.img3, y1, y2, x1, x2)
self.background = self.background[:, :, 0:3]
location_list.append([x1, y1, x2, y2])
# print('img3', self.img3.shape)
# print("x1,y1,x2,y2: ", x1, y1, x2, y2)
# sub4 img4
x_min, x_max, y_min, y_max = subarea_dict['sub4']
x1 = random.randint(x_min, x_max)
y1 = random.randint(y_min, y_max)
x2 = x1 + self.img4.shape[1]
y2 = y1 + self.img4.shape[0]
self.background = self.merge_img(self.background, self.img4, y1, y2, x1, x2)
self.background = self.background[:, :, 0:3]
location_list.append([x1, y1, x2, y2])
# print('img4', self.img4.shape)
# print("x1,y1,x2,y2: ", x1, y1, x2, y2)
#img5 random choice sub space:
x1, y1, x2, y2 = self.random_set_img(location_list, overlap_rate=0.1)
self.background = self.merge_img(self.background, self.img5, y1, y2, x1, x2)
self.background = self.background[:, :, 0:3]
location_list.append([x1, y1, x2, y2])
# 考虑一个img5把其他几个遮掉的问题
lis1,lis2,lis3,lis4,lis5 = location_list
if calculate_iou(lis1,lis5)>0.08 or calculate_iou(lis2,lis5)>0.08 \
or calculate_iou(lis3,lis5)>0.08 or calculate_iou(lis4,lis5)>0.08:
print('$%$%$%$%$%$%$%$%$%$%$%$%$%$%$%$%$%$%$%$%$%$%$%$%$%$%$%$%$')
continue
filename = background_prefix + '_' + str(100000 + self.i)
image_path = os.path.join(self.image_save_path, filename + '.jpg')
gray_img = cv2.cvtColor(self.background, cv2.COLOR_RGB2GRAY)
gray_img = gray_img[:, :, np.newaxis]
gray_img = np.dstack((gray_img, gray_img, gray_img))
cv2.imwrite(image_path, gray_img)
#############################################################
# # 获取原始图像的高度和宽度
# height, width = self.background.shape[:2]
# # 计算进行裁剪后的图像的新高度和新宽度,使其能够被3整除
# new_height = height - (height % 3)
# new_width = width - (width % 3)
# # 对图像进行裁剪
# self.background = self.background[:new_height, :new_width]
# cv2.imwrite(image_path,self.background)
# #LR
# image_path_lr = os.path.join(r'D:\pythonproject\generatedataset\test\lr', filename + '.jpg')
# lr_height = new_height // 3
# lr_width = new_width // 3
# resized_image = cv2.resize(self.background, (int(lr_width), int(lr_height)), interpolation=cv2.INTER_CUBIC)
# cv2.imwrite(image_path_lr, resized_image)
############################################################
# self.voc_annotation(self.img_label, filename, location_list)
self.coco_annotation(filename,location_list)
"================= test ===================="
# cv2.imshow('background',self.background)
# cv2.imshow('test', self.test)
# cv2.waitKey(0)
"============================================="
self.i += 1
if __name__ == "__main__":
dir_list = [r'D:\pythonproject\generatedataset\cross_target', r'D:\pythonproject\generatedataset\t_target']
label_list = ['cross_target', 't_target']
json_file_dir = 'D:\pythonproject\generatedataset\json'
generate_dataset = generate_dataset(dir_list, label_list, 'background', r'D:\pythonproject\generatedataset',
'VOC2007_C/JPEGImages_coco',
'VOC2007_C/Annotations_coco',
json_file_dir)
# target_dir_path_list,label_list,background_dir_path,project_abspath,image_save_path,annotation_save_path,json_dir_path:
generate_dataset.main()
# print(dict)