hn18001 / E2E-MLT

E2E-MLT - an Unconstrained End-to-End Method for Multi-Language Scene Text

E2E-MLT

E2E-MLT - an Unconstrained End-to-End Method for Multi-Language Scene Text code base for: https://arxiv.org/abs/1801.09919

@article{buvsta2018e2e,
  title={E2E-MLT-an unconstrained end-to-end method for multi-language scene text},
  author={Bu{\v{s}}ta, Michal and Patel, Yash and Matas, Jiri},
  journal={arXiv preprint arXiv:1801.09919},
  year={2018}
}

Requirements

• python3.x with
• opencv-python
• pytorch 0.4.1
• torchvision
• warp-ctc (https://github.com/SeanNaren/warp-ctc/)

Pretrained Models

e2e-mlt

wget http://ptak.felk.cvut.cz/public_datasets/SyntText/e2e-mlt.h5

Running Demo

python3 demo.py -model=e2e-mlt.h5

Data

• ICDAR MLT Dataset
• ICDAR 2015 Dataset
• RCTW-17
• Synthetic MLT Data (Arabic, Bangla, Chinese, Japanese, Korean, Latin )

Synthetic text has been generated using Synthetic Data for Text Localisation in Natural Images, with minor changes for Arabic and Bangla script rendering.

What we have found useful:

• for generating Arabic Scene Text: https://github.com/mpcabd/python-arabic-reshaper
• for generating Bangla Scene Text: PyQt4
• having somebody who can read non-latin scripts: we would like to thank Ali Anas for reviewing generated Arabic scene text.

Trainnig

python3 train.py -train_list=sample_train_data/MLT/trainMLT.txt -batch_size=8 -num_readers=5 -debug=0 -input_size=512 -ocr_batch_size=256 -ocr_feed_list=sample_train_data/MLT_CROPS/gt.txt

Acknowledgments

Code borrows from EAST and DeepTextSpotter