A Multi-Lingual OCR using CTC

---

This repository contains only the OCR unit that is used in this repository.

Please use Anaconda or miniconda for installation.

To run this model, you would need the following steps:

• conda env create -f environment.yml
• wget http://ptak.felk.cvut.cz/public_datasets/SyntText/e2e-mlt.h5
• conda activate ocr

OR

simply run:

• bash setup.sh
• conda activate ocr

make sure that you have a GPU

now you have 2 choices.

Run the OCR on images present in input_data and save the output in output_data:

bash start.sh

Please Note: the format in which the recognition result is saved is:

<image_name>_<recogintion_result>.png

Example:

if your image name is: img_1.jpg, and your recognition result is: hello_world.

The output image name would be: img_1_hello_world.png.

how to run images present in some random <input_image_path> and store output in some random <output_image_path>?

to do this instead of bash start.sh run: python eval.py -input-path=your_random_image_path -output_path=your_random_image_path

Reference:

@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}
}

About me: Go to http:www.pinakinathc.me