1. Introduction

This project includes the text detection source code and trained model about the word stroke region and text center block.

2. Installation

• Clone the repo

git clone git@github.com:zdlcaffe/WSRTCB.git  

• Then you can do as follow:

cd ${WSRTCB_root/Train_WSR_TCB/caffe}  
make –j  
make pycaffe 

3. Testing

3.1 Generate WSR/TCB score map

• Download the TD_MKEI_Word.caffemodel, trained on KAIT dataset.
• Then you can do as follow:

cd ${WSRTCB_root/}  
mkdir snapshot  
mkdir pre_model

• put TD_MKEI_Word.caffemodel to the fold of ${WSRTCB_root/Train_WSR_TCB/ snapshot}.

• Suppose you have downloaded the MSRA-TD500 dataset, execute the following commands to test the model on MSRA-TD500. Then you can do as follow:

cd ${WSRTCB_root/Train_WSR_TCB/demo}  
python Demo.py  

3.2 There are some samples:

3.3 Threshold WSR/TCB maps:

You can do as follow:

cd ${WSRTCB_root/Text_Demo}  
python fuse_thred	.py  

3.4 Generate detection results

You can do as follow:

cd ${WSRTCB_root/Text_Demo}  
python Demo_region_word.py

3.5 There are some samples:

4. Training

Download the pretrained model vgg16convs.caffemodel, and put it to ${WSRTCB_root/Train_WSR_TCB/pre_model}

4.1 Generate your map

Scripts for generating ground truth have been provided in the label_generate directory. It not hard to write a converting script for your own dataset.

4.2 Train your own model

Modify ${WSRTCB_root/Train_WSR_TCB/model/TD_MKEI_Word.py} to configure your dataset name and dataset path like:
......
data_params['root'] = 'data/MKEIWord'
data_params['source'] = "MKEI_Word.lst"
......

4.3 Start training

You can do as follow:

cd ${WSRTCB_root/Train_WSR_TCB}  
sh ./train.sh 

Citation

Use this bibtex to cite this repository:

@article{liu2019scene,
  title={Scene text detection with fully convolutional neural networks},
  author={Liu, Zhandong and Zhou, Wengang and Li, Houqiang},
  journal={Multimedia Tools and Applications},
  pages={1--23},
  year={2019},
  publisher={Springer}
}

Acknowlegement