zhangxiao339 / DenseCTC-Keras

chinese ocr, crnn, densectc, for rec text, 中英文文字识别OCR

中文

Introduction

The basic model of text recognition, based on densenet+ CTC, does not include LSTM, can take into account the recognition rate and performance, CPU on average 5ms on macbookpro i7
Based on Tensorflow and Keras, the end - to - end detection and recognition of variable length Chinese characters are realized

• text detect：DB
• text rec：DenseNet + CTC

Environment

tensorflow 1.13.1, keras 2.2.4

demo

train

1. Data

* 先处理数据集为统一大小
* 配置config

2. Training

* python app.py --mode trainval --gpu 'you gpu id'
* launch tensorboard

test

• got the text picture
• python app.py --mode test --image_path 'you image file' --gpu 'you gpu id'

reference

[1] https://github.com/chineseocr/chinese-ocr

[2] https://github.com/YCG09/chinese_ocr