use CNN to recognize charset
first you should prepare your image in 'train' dir,and the image name should like '0_A.jpg','1_B.jpg'
then you should prepare your image in 'test' dir,image name should like train image
after prepared image,you should edit the config.py to change image width and other things
at last,run the run.py
1.move trained model's weight file to pretrained dir
2.changeg config.py pretrained_path
3.use run.py's function of label_img() to label origin image