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
tensorflow 1.13.1, keras 2.2.4
* 先处理数据集为统一大小
* 配置config
* python app.py --mode trainval --gpu 'you gpu id'
* launch tensorboard
- got the text picture
- python app.py --mode test --image_path 'you image file' --gpu 'you gpu id'