CBNet: A Plug-and-Play Network for Segmentation-based Scene Text Detection

Official PyTorch implementation of "CBNet: A Plug-and-Play Network for Segmentation-based Scene Text Detection" in International Journal of Computer Vision (IJCV) 2024.

News

• (2024/03/21) Code is released.

Installation

First, clone the repository locally:

git clone https://github.com/XiiZhao/cbn.pytorch

Then, install PyTorch 1.4.0+, torchvision 0.5.0+, and other requirements:

conda install pytorch torchvision -c pytorch
pip install -r requirement.txt

Finally, for CPP type codes of post-processing:

# build cpp boundary-guided algorithms
cd models/post_processing/bg/
# prepare packages (e.g. pybind11, clipper), set Makefile and compile
make -j12

Dataset

Please refer to dataset/README.md for dataset preparation.

Training

CUDA_VISIBLE_DEVICES=0,1,2,3 python train.py ${CONFIG_FILE}

For example:

CUDA_VISIBLE_DEVICES=0,1,2,3 python train.py config/cbn/cbn_r18_ctw.py

Testing

Evaluate the performance

# single model type:
python test.py ${CONFIG_FILE} ${CHECKPOINT_FILE}
cd eval/
./eval_{DATASET}.sh

# batch modle type:
1) save all model path to a file (e.g. ctw_model.list)
2) sh batchEval.sh ctw_model.list

For example:

python test.py config/cbn/cbn_r18_ctw.py checkpoints/cbn_r18_ctw/checkpoint.pth.tar
cd eval/
./eval_ctw.sh

Evaluate the speed

python test.py ${CONFIG_FILE} ${CHECKPOINT_FILE} --report_speed

For example:

python test.py config/cbn/cbn_r18_ctw.py checkpoints/cbn_r18_ctw/checkpoint.pth.tar --report_speed

Acknowledgement

This repository is built using the PAN repository, thanks their wonderful work.

Citation

If you find our work is useful for your research, please kindly cite our paper.

@article{zhao2024cbnet,
  title={CBNet: A Plug-and-Play Network for Segmentation-based Scene Text Detection},
  author={Zhao, Xi and Feng, Wei and Zhang, Zheng and Lv, Jingjing and Zhu, Xin and Lin, Zhangang and Hu, Jinghe and Shao, Jingping},
  journal={International Journal of Computer Vision},
  pages={1--20},
  year={2024},
  publisher={Springer}
}

Contact

If you have any questions, you can submit your issue in this repository, or contact me from the email: [email protected]