The purpose of this project is to provide a set of easily modified character recognition model based on PyTorch framework.

At present, the code is still under adjustment, and it will take some time for the pre-training model to be released.

Basicly, I want the code to be clean and easy to read. Some utilities you can add by yourself.

This implementation ennable you to train large amout of imgae-text data on gpu. The imgae transformation and the decoder is adapted to gpu too. So the rate of gpu utilization is improved greatly.

Hope you can enjoy your training.

you can use the datasets from aster.pytorch (link password: wi05).

or you can use the datasets from deep-text-recognition-benchmark (link password: rryk)(much more dataset is provided on google drive on this repo).

or you can create your own datasets as long as follow the same format.

python train.py

python test.py

aster.pytorch

@inproceedings{wang2020scene,
  title={Scene Text Recognition With Linear Constrained Rectification},
  author={Gang, Wang and Huaping, Zhang and jianyun, Shang},
  booktitle={2020 International Conference on Computational Science and Computational Intelligence (CSCI)},
  year={2020},
  organization={IEEE}
}