# -*- coding: utf-8 -*-
"""
@from:www.linerect.com
@author: duimu
@contact: duimu@qq.com
@Created on: 2019/12/17 10:34
本demo完全根据mjsynth 数据集进行测试，如果需要根据自己的图片生成，可以按照实例操作
第一步：
"""
import tensorflow as tf
import numpy as np
import os
from Step0_CrnnDefines import *
#根据图片生成所需要的txt文件，该txt文件用来生成tfrecords数据
#按照文件路径 lable的形式存储
def GetCharDic(picPath,writeTrainPath,writeValPath,radiu=0.2):
"""
:param picPath: 图片目录 eg: d:/dataset/images
:param writeTrainPath: 生成train.txt的详细路径 eg: d:/dataset/train.txt
:param writeValPath: 生成val.txt详细路径 eg: d:/dataset/val.txt
:param radiu: 验证数据占所有数据中的比率
:return:
"""
nameinfo=[]
for root, dirs, files in os.walk(picPath):
for f in files:
info = os.path.join(root, f)
info=info[len(picPath)+1:]
f = f.split('_')
info=info+' '
for char in f[1]:
if char >='a'and char <='z':
char=str(ord(char)-ord('a')+1)
elif char >='A'and char <='Z':
char = str(ord(char) - ord('A')+1)
elif char >='0'and char <='9':
char = str(ord(char) - ord('0')+27)
else:
continue
info=info+char+' '
nameinfo.append(info)
if len(nameinfo)%100==0:
print('size:',len(nameinfo))
trainfile = open(writeTrainPath, 'w')
valfile= open(writeValPath, 'w')
totalcount=len(nameinfo)
count=0
valCount=int(totalcount * (1.0 - radiu))
for line in nameinfo:
count=count+1
if count<valCount:
trainfile.write(line)
trainfile.write('\n')
elif count>=valCount:
valfile.write(line)
valfile.write('\n')
trainfile.close()
valfile.close()
if __name__ == '__main__':
GetCharDic('F:/0_dataset/mjsynth/set',
'F:/0_dataset/mjsynth/train1.txt',
'F:/0_dataset/mjsynth/val1.txt',
0.2)
