def main(db_fname):
print('START!')
db = h5py.File(db_fname, 'r')
dsets = sorted(db['data'].keys())
print ("total number of images : ", colorize(Color.RED, len(dsets), highlight=True))
print()
for num ,k in enumerate(dsets):
#add txt
f = open('{}/gt_img_{}.txt'.format(GT_PATH,num),'w')
rgb = db['data'][k][...]
charBB = db['data'][k].attrs['charBB']
wordBB = db['data'][k].attrs['wordBB']
txt = db['data'][k].attrs['txt']
#print('wordBB:',wordBB)
#print('txt:',txt)
#print('k',k)
#f.write()
viz_textbb(rgb, [charBB], wordBB, num=num, index=k, txt=txt)
f.close()
print( "image name : ", colorize(Color.RED, k, bold=True))
#print (" ** no. of chars : ", colorize(Color.YELLOW, charBB.shape[-1]))
print (" ** no. of words : ", colorize(Color.YELLOW, wordBB.shape[-1]))
print (" ** text : ", colorize(Color.GREEN, txt))
# print " ** text : ", colorize(Color.GREEN, txt.encode('utf-8'))
# if 'q' in raw_input("next? ('q' to exit) : "):
# break
db.close()1. Taget
text.py
for i_idx, image in enumerate(self.images_root.glob('*.jpg')):
# image = pathlib.Path('/data/ocr/det/icdar2015/detection/train/imgs/img_756.jpg')
# gt = pathlib.Path('/data/ocr/det/icdar2015/detection/train/gt/gt_img_756.txt')
gt = self.gt_root / image.with_name('gt_{}'.format(image.stem)).with_suffix('.txt').name
with gt.open(mode='r') as f:
bboxes = []
texts = []
for line in f:
text = line.strip('\ufeff').strip('\xef\xbb\xbf').strip().split(',')
x1, y1, x2, y2, x3, y3, x4, y4 = list(map(float, text[:8]))
bbox = [[x1, y1], [x3, y3]]
transcript = text[8]
if transcript == '###' and self.training:
continue
bboxes.append({"image_id":i_idx, "category_id":1, "bbox":bbox}) #label 1 for text, 0 for background
texts.append(transcript)
if len(bboxes) > 0:
bboxes = np.array(bboxes)
all_bboxs.append(bboxes)
all_texts.append(texts)
all_images.append(image)
all_ids.append(i_idx)
return all_images, all_bboxs, all_texts, all_ids
sample gt
data / tain / gt , imgs
SAMPLE
gt_img_1.txt
377,117,463,117,465,130,378,130,Genaxis Theatre
493,115,519,115,519,131,493,131,[06]
374,155,409,155,409,170,374,170,###
492,151,551,151,551,170,492,170,62-03
376,198,422,198,422,212,376,212,Carpark
494,190,539,189,539,205,494,206,###
374,1,494,0,492,85,372,86,###
2. SynthText - kr
def main(db_fname):
db = h5py.File(db_fname, 'r')
dsets = sorted(db['data'].keys())
print ("total number of images : ", colorize(Color.RED, len(dsets), highlight=True))
print()
for k in dsets:
rgb = db['data'][k][...]
charBB = db['data'][k].attrs['charBB']
wordBB = db['data'][k].attrs['wordBB']
txt = db['data'][k].attrs['txt']
viz_textbb(rgb, [charBB], wordBB, index=k, txt=txt)
print( "image name : ", colorize(Color.RED, k, bold=True))
print (" ** no. of chars : ", colorize(Color.YELLOW, charBB.shape[-1]))
print (" ** no. of words : ", colorize(Color.YELLOW, wordBB.shape[-1]))
print (" ** text : ", colorize(Color.GREEN, txt))h5 file path : gen/dataset_sample.h5
total number of images : 1
wordBB: [[[208.6069 133.4994 101.2383 84.159134 93.3638 ]
[395.68124 539.3367 182.21889 198.9645 172.35709 ]
[393.8134 539.36487 181.0756 197.66335 170.69101 ]
[206.73904 133.5276 100.09502 82.85799 91.69772 ]]
[[378.6737 426.93433 247.17474 97.715576 178.44614 ]
[385.53638 426.81396 249.13928 101.58473 180.8259 ]
[436.45404 521.8998 296.2671 140.1925 236.12885 ]
[429.59137 522.02014 294.30255 136.32335 233.74908 ]]]
txt: ['헤이브릴' '충저우해협' '된지' '토반묘' '삼작']
k ant+hill_10.jpg_0377,117,463,117,465,130,378,130,Genaxis Theatre
493,115,519,115,519,131,493,131,[06]
374,155,409,155,409,170,374,170,###
492,151,551,151,551,170,492,170,62-03
376,198,422,198,422,212,376,212,Carpark
494,190,539,189,539,205,494,206,###
374,1,494,0,492,85,372,86,###visualize_results.py
# split words and create jpg files
def _tojpg(text_im, polygon, index, k, typ):
img = Image.fromarray(text_im, 'RGB')
img.save(IMG_PATH+'/image/'+k+'.jpg')
# convert to numpy (for convenience)
imArray = np.asarray(img)
maskIm = Image.new('L', (imArray.shape[1], imArray.shape[0]), 0)
ImageDraw.Draw(maskIm).polygon(polygon, outline=1, fill=1)
mask = np.array(maskIm)
# assemble new image (uint8: 0-255)
newImArray = np.empty(imArray.shape,dtype='uint8')
# colors (three first columns, RGB)
newImArray[:,:,:3] = imArray[:,:,:3]
# back to Image from numpy
newIm = Image.fromarray(newImArray, "RGB")
x_min = min([x[0] for x in polygon])
x_max = max([x[0] for x in polygon])
y_min = min([x[1] for x in polygon])
y_max = max([x[1] for x in polygon])
newIm = newIm.crop((x_min, y_min, x_max, y_max))
if typ == "char":
newIm.save(IMG_PATH+"/characters/"+typ+"_"+str(index)+".jpg")
f1 = open('ch_coords.txt', 'a')
f1.write(typ+str(index)+".jpg"+"\t"+str(x_min)+","+str(y_min)+","+str(x_max)+","+str(y_max)+"\n")
elif typ == "word":
newIm.save(IMG_PATH+"/words/"+typ+"_"+str(index)+".jpg")
f2 = open('wd_coords.txt', 'a')
f2.write(typ+str(index)+".jpg"+"\t"+str(x_min)+","+str(y_min)+","+str(x_max)+","+str(y_max)+"\n")visualize_result.py
def viz_textbb(fignum,text_im, bb_list,alpha=1.0):
"""
text_im : image containing text
bb_list : list of 2x4xn_i boundinb-box matrices
"""
plt.close(fignum)
plt.figure(fignum)
plt.imshow(text_im)
#plt.hold(True)
H,W = text_im.shape[:2]
for i in range(len(bb_list)):
bbs = bb_list[i]
ni = bbs.shape[-1]
for j in range(ni):
bb = bbs[:,:,j]
bb = np.c_[bb,bb[:,0]]
plt.plot(bb[0,:], bb[1,:], 'r', linewidth=2, alpha=alpha)
plt.gca().set_xlim([0,W-1])
plt.gca().set_ylim([H-1,0])
plt.show(block=False)gt_ synthgen.py
#def coords 에서 tmp 아래와 같이 출력
#tmp=list(zip(x_list, y_list))
[(235, 29), (299, 33), (297, 62), (232, 58)]
[(230, 70), (473, 121), (453, 215), (210, 164)]
[(129, 445), (244, 445), (244, 508), (129, 508)]
[(135, 10), (210, 10), (210, 41), (135, 41)]
[(483, 5), (586, 11), (582, 78), (480, 73)]
[(57, 499), (123, 485), (128, 510), (62, 523)]
[(217, 6), (279, 6), (279, 34), (217, 33)]
[(209, 181), (372, 199), (367, 247), (203, 229)]
[(4, 67), (115, 68), (115, 131), (4, 130)]
[(475, 84), (549, 87), (548, 126), (473, 123)]
[(5, 481), (83, 466), (88, 492), (10, 506)]
[(369, 220), (424, 226), (421, 253), (366, 247)]
[(73, 274), (199, 274), (199, 322), (73, 321)]
[(500, 129), (594, 132), (593, 163), (499, 159)]
[(4, 454), (94, 436), (98, 459), (9, 476)]
[(335, 267), (395, 287), (387, 314), (326, 293)]
[(147, 319), (188, 320), (188, 341), (147, 341)]
[(423, 10), (483, 13), (481, 52), (420, 49)]
[(298, 46), (334, 48), (334, 73), (297, 72)]
[(409, 192), (454, 197), (451, 220), (406, 214)]
[(277, 398), (522, 399), (521, 509), (276, 508)]
[(548, 73), (591, 75), (589, 113), (546, 111)]
f = open('{}/gt_img_{}.txt'.format(GT_PATH,num),'w')
IMG_PATH ='/content/drive/Shareddrives/OCR_Transformer/SynthText_kr/imgs'
GT_PATH = '/content/drive/Shareddrives/OCR_Transformer/SynthText_kr/gt'
h5_PATH ='gen/dataset_sample.h5' #'gen/dataset_kr.h5'RESULT (visualize_result.py)
"""
Visualize the generated localization synthetic
data stored in h5 data-bases
"""
from __future__ import division
import os
import os.path as osp
import numpy as np
import matplotlib
matplotlib.use('Agg')
import matplotlib.pyplot as plt
import h5py
import cv2
from common import *
import itertools
from PIL import Image, ImageDraw
import re
import json
# initialize index
num = 1
char_ind = 0
word_ind = 0
def viz_textbb(text_im, charBB_list, wordBB, num, index, txt, alpha=1.0):
"""
text_im : image containing text
charBB_list : list of 2x4xn_i bounding-box matrices
wordBB : 2x4xm matrix of word coordinates
"""
#plt.close(1)
#plt.figure(1)
#plt.imshow(text_im)
img = Image.fromarray(text_im, 'RGB')
img.save('{}/img_{}.jpg'.format(IMG_PATH,num))
#plt.hold(True)
H,W = text_im.shape[:2]
# plot the word-BB:
txt_index = 0 #txt 순서대로 출력
for i in range(wordBB.shape[-1]):
bb = wordBB[:,:,i]
bb = np.c_[bb,bb[:,0]]
#plt.plot(bb[0,:], bb[1,:], 'g', alpha=alpha)
coords(bb[0,:], bb[1,:], num, index, text_im, txt[txt_index])
txt_index = txt_index +1
#_tojpg(text_im, bb, index, num, index, typ='word')
# visualize the indiv vertices:
#vcol = ['r','g','b','k']
#for j in range(4):
# plt.scatter(bb[0,j],bb[1,j],color=vcol[j])
# plot the character-BB:
#for i in range(len(charBB_list)):
# bbs = charBB_list[i]
# ni = bbs.shape[-1]
# for j in range(ni):
# bb = bbs[:,:,j]
# bb = np.c_[bb,bb[:,0]]
# plt.plot(bb[0,:], bb[1,:], 'r', alpha=alpha/2)
# coords(bb[0,:], bb[1,:], index, text_im, "char")
# generate image files and coordinate files
def coords(x_list, y_list, num, k, img, word):
tmp = []
global char_ind
global word_ind
# remove alpha
x_list=x_list[:-1]
y_list=y_list[:-1]
# change datatype
x_list=[int(x) for x in x_list]
y_list=[int(y) for y in y_list]
# generate coords list
tmp = list(zip(x_list, y_list))
with open('{}/gt_img_{}.txt'.format(GT_PATH,num),'a') as f :
for i,xy in enumerate(tmp) :
x,y = xy
f.write('{},{},'.format(x,y))
if (i+1) % 4 == 0 :
f.write('{}\n'.format(word) )
#coords to image file
k=k.split('.')[0]
word_ind = word_ind + 1
#_tojpg(img, tmp, word_ind, num,k, word)
# split words and create jpg files
def _tojpg(text_im, polygon, index, num, k, word):
img = Image.fromarray(text_im, 'RGB')
img.save(IMG_PATH+'/img_{}.jpg'.format(num))
# convert to numpy (for convenience)
imArray = np.asarray(img)
maskIm = Image.new('L', (imArray.shape[1], imArray.shape[0]), 0)
ImageDraw.Draw(maskIm).polygon(polygon, outline=1, fill=1)
mask = np.array(maskIm)
# assemble new image (uint8: 0-255)
newImArray = np.empty(imArray.shape,dtype='uint8')
# colors (three first columns, RGB)
newImArray[:,:,:3] = imArray[:,:,:3]
# back to Image from numpy
newIm = Image.fromarray(newImArray, "RGB")
x_min = min([x[0] for x in polygon])
x_max = max([x[0] for x in polygon])
y_min = min([x[1] for x in polygon])
y_max = max([x[1] for x in polygon])
#print('**')
#print(typ+str(index)+".jpg"+"\t"+str(x_min)+","+str(y_min)+","+str(x_max)+","+str(y_max)+"\n")
newIm = newIm.crop((x_min, y_min, x_max, y_max))
newIm.save(IMG_PATH+"/words/"+"testtt_"+str(index)+".jpg") #save image
f2 = open('wd_coords.txt', 'a')
f2.write(typ+str(index)+".jpg"+"\t"+str(x_min)+","+str(y_min)+","+str(x_max)+","+str(y_max)+"\n")
def main(db_fname):
db = h5py.File(db_fname, 'r')
dsets = sorted(db['data'].keys())
print ("total number of images : ", colorize(Color.RED, len(dsets), highlight=True))
print()
for num ,k in enumerate(dsets):
#add txt
f = open('{}/gt_img_{}.txt'.format(GT_PATH,num),'w')
rgb = db['data'][k][...]
charBB = db['data'][k].attrs['charBB']
wordBB = db['data'][k].attrs['wordBB']
txt = db['data'][k].attrs['txt']
#print('wordBB:',wordBB)
#print('txt:',txt)
#print('k',k)
#f.write()
viz_textbb(rgb, [charBB], wordBB, num=num, index=k, txt=txt)
f.close()
print( "image name : ", colorize(Color.RED, k, bold=True))
#print (" ** no. of chars : ", colorize(Color.YELLOW, charBB.shape[-1]))
print (" ** no. of words : ", colorize(Color.YELLOW, wordBB.shape[-1]))
print (" ** text : ", colorize(Color.GREEN, txt))
# print " ** text : ", colorize(Color.GREEN, txt.encode('utf-8'))
# if 'q' in raw_input("next? ('q' to exit) : "):
# break
db.close()
if __name__=='__main__':
IMG_PATH ='/content/drive/Shareddrives/OCR_Transformer/SynthText_kr/imgs'
GT_PATH = '/content/drive/Shareddrives/OCR_Transformer/SynthText_kr/gt'
h5_PATH ='gen/dataset_kr.h5' #'gen/dataset_kr.h5'
if os.path.isdir(IMG_PATH) == False :
os.mkdir(IMG_PATH)
if os.path.isdir(GT_PATH) == False :
os.mkdir(GT_PATH)
print('h5 file path : ',h5_PATH)
main(h5_PATH)3. SynthText - eng (original)
+ char_freq file 연결
(나머지 동일)
4. Koean _aihub
'Project > OCR' 카테고리의 다른 글
| [Project] E2E_OCR Model Architecture (0) | 2022.03.01 |
|---|---|
| [CODE] SynthText - Korean ver 응용 (0) | 2022.02.15 |
