File size: 12,121 Bytes
2da3364 |
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 |
# pip install numpy==1.26.4 opencv-python==4.6.0.66
# see doc\lang\programming\pytorch\文本检测\DBNET 论文代码都有
"""
原图每一行剪出一张图, 对这张图做字符级标注
给 DBNet 官方代码用
将阿里OCR 的识别结果(图片和标注)转换成 icdar2015 格式 (注意:它的文本是含 utf8 bom 的)
"""
"""
icdar2015 文本检测数据集
标注格式: x1,y1,x2,y2,x3,y3,x4,y4,text
其中, x1,y1为左上角坐标,x2,y2为右上角坐标,x3,y3为右下角坐标,x4,y4为左下角坐标。
### 表示text难以辨认。
"""
import random
from pathlib import Path
import os
import glob
import base64
from importlib.resources import path
import math
import numpy as np
import cv2
import json
import decimal
import datetime
from pickletools import uint8
class DecimalEncoder(json.JSONEncoder):
def default(self, o):
if isinstance(o, decimal.Decimal):
return float(o)
elif isinstance(o, datetime.datetime):
return str(o)
super(DecimalEncoder, self).default(o)
def save_json(filename, dics):
with open(filename, 'w', encoding='utf-8') as fp:
json.dump(dics, fp, indent=4, cls=DecimalEncoder, ensure_ascii=False)
fp.close()
def load_json(filename):
with open(filename, encoding='utf-8') as fp:
js = json.load(fp)
fp.close()
return js
# convert string to json
def parse(s):
return json.loads(s, strict=False)
# convert dict to string
def string(d):
return json.dumps(d, cls=DecimalEncoder, ensure_ascii=False)
def transform(points, M):
# points 算出四个点变换后移动到哪里了
# points = np.array([[word_x, word_y], # 左上
# [word_x + word_width, word_y], # 右上
# [word_x + word_width, word_y + word_height], # 右下
# [word_x, word_y + word_height], # 左下
# ])
# add ones
ones = np.ones(shape=(len(points), 1))
points_ones = np.hstack([points, ones])
# transform points
transformed_points = M.dot(points_ones.T).T
transformed_points_int = np.round(
transformed_points, decimals=0).astype(np.int32) # 批量四舍五入
return transformed_points_int
def cutPoly(img, pts):
# img = cv2.imdecode(np.fromfile('./t.png', dtype=np.uint8), -1)
# pts = np.array([[10,150],[150,100],[300,150],[350,100],[310,20],[35,10]])
## (1) Crop the bounding rect
rect = cv2.boundingRect(pts)
x,y,w,h = rect
croped = img[y:y+h, x:x+w].copy()
## (2) make mask
pts = pts - pts.min(axis=0)
mask = np.zeros(croped.shape[:2], np.uint8)
cv2.drawContours(mask, [pts], -1, (255, 255, 255), -1, cv2.LINE_AA)
## (3) do bit-op
dst = cv2.bitwise_and(croped, croped, mask=mask)
## (4) add the white background
bg = np.ones_like(croped, np.uint8)*255
cv2.bitwise_not(bg,bg, mask=mask)
dst2 = bg+ dst
# cv2.imwrite("croped.png", croped)
# cv2.imwrite("mask.png", mask)
# cv2.imwrite("dst.png", dst)
# cv2.imwrite("dst2.png", dst2)
return dst2
if __name__ == "__main__":
# 验证原版的文本标记框
# im = './datasets/icdar2015/train_images/img_1.jpg'
# gt = './datasets/icdar2015/train_gts/gt_img_1.txt'
# 验证自已生成的标记框
im = './icdar2015_aliocr_char/train_images/img_00000001.jpg'
gt = './icdar2015_aliocr_char/train_gts/gt_img_00000001.txt'
if os.path.exists(gt):
items = []
reader = open(gt, 'r', encoding='utf-8-sig').readlines()
for line in reader:
item = {}
parts = line.strip().split(',')
label = parts[-1]
if 'TD' in gt and label == '1':
label = '###'
line = [i.strip('\ufeff').strip('\xef\xbb\xbf') for i in parts]
if 'icdar' in gt:
poly = np.array(list(map(float, line[:8]))).reshape(
(-1, 2)).tolist()
else:
num_points = math.floor((len(line) - 1) / 2) * 2
poly = np.array(list(map(float, line[:num_points]))).reshape(
(-1, 2)).tolist()
item['poly'] = poly
item['text'] = label
# 多边形是用一个个的点表示的,起点连接第二个点,第二个连接第三个 ... 最后一点连接起点,构成一个闭合的区域
item['points'] = poly
# 此标记表示文字模糊不可辨认,文本框的标记是不可靠的
item['ignore'] = True if label == '###' else False
items.append(item)
img = cv2.imdecode(np.fromfile(im, dtype=np.uint8), -1)
# DBNet 原版代码只能处理彩图,所以统一处理成彩图
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
for i in range(len(items)):
poly = items[i]['poly']
poly = np.array(poly)
poly = poly.astype(np.int32)
# cv2.fillPoly(img, pts=[ poly ], color=(0, 0, 255))
b = random.randint(0, 255) # 用来生成[a,b]之间的随意整数,包括两个边界值。
g = random.randint(0, 255)
r = random.randint(0, 255)
# 只画线,不填充 # 就是画线,从起点连到第二个点 ... 最后一个点连到第一个点
cv2.polylines(img, [poly], isClosed=True,
color=(b, g, r), thickness=1)
cv2.imwrite("poly.jpg", img)
# cv2.imshow("poly", img)
# cv2.waitKey()
# 开始转换
out_dir = 'icdar2015_aliocr_char'
if os.path.exists(out_dir):
import shutil
shutil.rmtree(out_dir)
# https://help.aliyun.com/document_detail/294540.html 阿里云ocr结果字段定义
# prism-wordsInfo 里的 angle 文字块的角度,这个角度只影响width和height,当角度为-90、90、-270、270,width和height的值需要自行互换
dir_json = './data/json' # '/yingedu/www/ocr_server/data/json'
dir_img = './data/img' # '/yingedu/www/ocr_server/data/img'
train_list = []
train_list_path = os.path.join(out_dir, 'train_list.txt')
test_list = []
test_list_path = os.path.join(out_dir, 'test_list.txt')
g_count = 1
count = 1
json_paths = glob.glob('{}/*.json'.format(dir_json), recursive=True)
for json_path in json_paths:
base = Path(json_path).stem
img_train_path = os.path.join(dir_img, '{}.txt'.format(base))
if not os.path.exists(img_train_path): # 没有相应的图片,可能被删除了
continue
jsn = load_json(json_path)
with open(img_train_path, "r", encoding="utf-8") as fp:
imgdata = fp.read()
imgdata = base64.b64decode(imgdata)
imgdata = np.frombuffer(imgdata, np.uint8)
img = cv2.imdecode(imgdata, cv2.IMREAD_UNCHANGED)
# cv2.imshow('img', img)
# cv2.waitKey(0)
if len(img.shape) != 3: # 转彩图
img_color = cv2.cvtColor(img, cv2.COLOR_GRAY2BGR)
img = cv2.cvtColor(img, cv2.COLOR_GRAY2BGR) # DBNet 原版只能处理彩图,这里转一下
else:
img_color = img.copy()
img_color_origin = img_color.copy()
img_color_origin2 = img_color.copy()
wordsInfo = jsn['prism_wordsInfo']
for j in range(len(wordsInfo)):
jo = wordsInfo[j]
word = jo["word"]
charInfo = jo["charInfo"]
# prism-wordsInfo 里的 angle 文字块的角度,这个角度只影响width和height,当角度为-90、90、-270、270,width和height的值需要自行互换
angle = jo['angle']
img_color = img_color_origin.copy()
"""
x y 宽高全部不靠谱, pos 里是对的
"""
# 四个角的位置 # 左上、右上、右下、左下,当NeedRotate为true时,如果最外层的angle不为0,需要按照angle矫正图片后,坐标才准确
pos = jo["pos"]
x = int(pos[0]["x"]) # 左上
y = int(pos[0]["y"])
x2 = int(pos[2]["x"]) # 右下
y2 = int(pos[2]["y"])
lu = [pos[0]['x'], pos[0]['y']] # left up 四个角顺时针方向数
ru = [pos[1]['x'], pos[1]['y']]
rd = [pos[2]['x'], pos[2]['y']]
ld = [pos[3]['x'], pos[3]['y']]
min_x = min(lu[0], ld[0])
max_x = max(ru[0], rd[0])
min_y = min(lu[1], ru[1])
max_y = max(rd[1], ld[1])
rows, cols = img.shape[:2]
if max_y >= rows:
max_y = rows - 1
if max_x >= cols:
max_x = cols - 1
crop = img[min_y:max_y+1, min_x:max_x+1]
# cv2.imshow("crop", crop)
# cv2.waitKey()
is_train_img = random.choices([0, 1], weights=[0.15, 0.85])[0]
# 85% 的概率是训练图
img_name = "img_{:08d}.jpg".format(g_count)
gt_name = "gt_img_{:08d}.txt".format(g_count)
gt_txt_list = []
img_train_path = os.path.join(out_dir, 'train_images', img_name)
img_train_gt_path = os.path.join(out_dir, 'train_gts', gt_name)
img_test_path = os.path.join(out_dir, 'test_images', img_name)
img_test_gt_path = os.path.join(out_dir, 'test_gts', gt_name)
dir1 = os.path.dirname(img_train_path)
dir2 = os.path.dirname(img_train_gt_path)
dir3 = os.path.dirname(img_test_path)
dir4 = os.path.dirname(img_test_gt_path)
if not os.path.exists(dir1):
os.makedirs(dir1)
if not os.path.exists(dir2):
os.makedirs(dir2)
if not os.path.exists(dir3):
os.makedirs(dir3)
if not os.path.exists(dir4):
os.makedirs(dir4)
if is_train_img:
train_list.append(img_name)
cv2.imwrite(img_train_path, crop)
else:
test_list.append(img_name)
cv2.imwrite(img_test_path, crop)
for info in charInfo:
wd = info["word"]
wd_x = info["x"]
wd_y = info["y"]
wd_w = info["w"]
wd_h = info["h"]
wd_crop = img[wd_y:wd_y+wd_h, wd_x:wd_x+wd_w]
wd_x_local = wd_x - min_x
wd_y_local = wd_y - min_y
wd_crop2 = crop[wd_y_local:wd_y_local+wd_h, wd_x_local:wd_x_local+wd_w]
# cv2.imshow("wd_crop", wd_crop2)
# cv2.waitKey()
lu_wd = [wd_x_local, wd_y_local]
ru_wd = [wd_x_local+wd_w, wd_y_local]
rd_wd = [wd_x_local+wd_w, wd_y_local+wd_h]
ld_wd = [wd_x_local, wd_y_local+wd_h]
# 生成 icdar2015 格式的人工标记训练数据(用于训练官方DB)
gt_txt_list.append( "{},{},{},{},{},{},{},{},{}".format(lu_wd[0], lu_wd[1], ru_wd[0], ru_wd[1], rd_wd[0], rd_wd[1], ld_wd[0], ld_wd[1], wd) )
gt_txt = "\n".join(gt_txt_list)
if is_train_img:
with open(img_train_gt_path, 'w', encoding='utf-8') as f:
f.write(gt_txt)
else:
with open(img_test_gt_path, 'w', encoding='utf-8') as f:
f.write(gt_txt)
g_count += 1
print(f'### one task one. {count} / {len(json_paths)}')
count += 1
train_list_txt = "\n".join(train_list)
test_list_txt = "\n".join(test_list)
with open(os.path.join(out_dir, "train_list.txt"), 'w', encoding='utf-8') as f:
f.write(train_list_txt)
with open(os.path.join(out_dir, "test_list.txt"), 'w', encoding='utf-8') as f:
f.write(test_list_txt)
print('### all task done.')
|