/ocr_exp_v1

Experimental repository for Korean OCR

Primary LanguagePython

(1) Directory Explanation of the project ocr_exp_v1

  • The training process for the text detection, text recognition all uses the base_trainer.py in the tools folder, and the training entry code is the base_runner.py
ocr_exp_v1
|__ config
|__ flask_serve [Flask Demo App to show the Detection - Recognition Process]
|__ key_info_extraction [TODO]
|__ text_detection [CTPN Model + Data Utils + Loss Function]
|__ text_recognition [Hangul Net Model + Baseline Model]
|__ tools
    |- base_trainer.py
|- base_runner.py

(2) Settings

1. Clone the repository

git clone https://github.com/369-Speaking-Fridgey/ocr_exp_v1.git

2. Installation

  • First, you must create a virtual enviornment with th 3.6 version of python
  • Activate your enviornment
    • in the example below, you will be making an enviornment with the name venv
  • Move to the cloned repository
  • Install the required libraries with the requirenemts.txt file 
    conda create -n venv python=3.6
conda activate venv
cd ocr_exp_v1
pip install requirements.txt

(3) To run the Flask Demo App

--> This can be done after the SETTING is finished (=previous step)

0. Move to the folder flask_serve

cd ocr_exp_v1/flask_serve

1. Run the Flask Server in Local Host

  • Since the 5000 local host port is running for the MLFlow operation, 3000 Port should be used.
flask run -h localhost -p 3000

2. The main page of the demo app

3. Select the image you want to test on and check the results

  • The text recognition result is not yet entirely successful - the Model is still training and trained weights will be updated soon

(4) Training (Currently available for the CTPN Model)

1. Move to the ocr_exp_v1 folder

cd ocr_exp_v1

2. Check the ctpn_detect_config.yml file in the config folder

  • You are able to change the settings in the train_configuration part
  • The custom data to train on is available in the zip format
  • The data must be stored in another folder, and currently the dataset I used to train is not uploaded in the repository.
    1. Download the Image Zip Data from https://drive.google.com/drive/u/0/folders/1MIqs8PlNmuD3w2JZ91Mwvc9JlOm-pBGR here.
    2. Download the Label Zip Data from https://drive.google.com/file/d/1y4jwixUEfG4FSez-vJHRnv3C2-CwR4NH/view?usp=share_link here.
    3. Now, place the downloaded image & label data in the data folder outside the ocr_exp_v1 folder
      • So, your structure must look like
      root
|__ data [MUST MAKE THIS FOLDER]
    |_ zip datas downloaded (image & label)
|__ ocr_exp_v1
    4. After that, change the img_path and label_path in the configuration file
  • Other important settings for training the CTPN model is customizable with the detect_configuration in the ctpn_detect_config.yml file.

3. Train

python3 base_runner.py