Skip to content

Latest commit

 

History

History
36 lines (23 loc) · 1.07 KB

README.md

File metadata and controls

36 lines (23 loc) · 1.07 KB
Raw

CRNN-TF-simple-implemented

This is a simple tensorflow implementation of CRNN which can recognize Chinese.

Inspired by (& copied from) https://github.com/Belval/CRNN and https://github.com/watsonyanghx/CNN_LSTM_CTC_Tensorflow

Reqiurements

  • Python 3.4+
  • tensorflow 1.4+
  • pillow
  • scipy
  • numpy

Anaconda Recommonded.

Data prepare

The image files and the label files should have the same name. The label file is a csv file with the format in each line like this: bottom,left,right,top,value

Example label file:

bottom,left,right,top,value
109,137,342,59,中信银行
118,1110,1206,69,账单
239,50,172,217,科目： 20102

Put the image files and the label files in train/ dir

To run the inference procedure, put the image files and the annotation files in val/ dir. The annotation file should have the same format with the label file, except that there is no value entry at the end of each line.

Run the model

Simply run python3 main.py

In the main.py file, you can choose which mode to use, either "train" or "infer". Just modify the code.