OCR Labeling Tool

Introduction

This is a really simple web app to hand label text data for OCR tasks built on Flask and PyQt.

Installation

First clone the project:

git clone https://github.com/pooya-mohammadi/ocr-labeling-tool

Install the requirements:

cd ocr-labeling-tool # enter to root project
pip install -r requirements.txt

Configs & Settings

The initial configs are located directly in settings.py module. The explanation of the hyper-parameters are as follows:

• DATA_PATH: The folder which contains the images.
  • All the images should have the same naming convention which is as follows:
  • {index}_{text}.{ext}.
  • A sample can contain several indices split by "_"/underline
• RESULTS_PATH: The directory which the results are saved in after being labeled.

Architecture

First time you run the app, a file called ocr-labeling-cursor-{name}-cursor.json is created(the name is equal to data_path last folder name). This file is responsible to gather your images' paths and track your progress, so that every time you run this code you will continue where you left off. (Note: Due to my laziness, the cursor module does not track any changes made on disk to the source images folder, so one has to manually modify the cursor config file under images key or delete the whole thing!) This file tracks the index to read and all images paths in the source folder. A sample cursor.json file would be like below:

{
  "file_index_to_read": 1,
  "images": {
    "1": "true-soule.jpg",
    "2": "Eskillstuna.jpg",
    "3": "mistitles.jpg",
    "4": "UNLABELED_1.jpg",
    "5": "coochampion.jpg",
    "6": "UNLABELED_2.jpg",
    "7": "UNLABELED_3.jpg",
    "8": "ennomic.jpg",
    "9": "erytrocytosis.jpg",
    "10": "well-scented.jpg"
  }
}

• file_index_to_read: images are indexed like above. this parameter sets the index to start from when the app is run.
• images: index to image paths mapping

When the app is running, the images show up based on the ordering in cursor.json. By default, the app considers that your files are named after their text. (Probably by another OCR model like EasyOCR) Then you have three options:

1. Edit or write text you desire in the text box and hit Save. The file with the registered text + the index will be saved to the output folder. (If you hit Save without changing the default text that comes up, the same text will be registerd as the result)
2. Hit Skip to ignore the image and text all together so that nothing is saved to the output folder.
3. Choose a desired index from cursor.json and hit Jump. if the index is valid the file_index_to_read will be changed to that. Use this in case you need to modify a previously registered image to overwrite it. (Make sure to jump to your last index you were at)

Run

Simply run the python app:

$ python app.py
* Serving Flask app "app" (lazy loading)
 * Environment: production
   WARNING: This is a development server. Do not use it in a production deployment.
   Use a production WSGI server instead.
 * Debug mode: off
 * Running on http://127.0.0.1:8002/ (Press CTRL+C to quit)

Qt window will pop up but if you prefer web apps you can also navigate to http://localhost:8002

References:

1. https://github.com/pooya-mohammadi/deep_utils
2. https://github.com/arxyzan/ocr-labeling-tool