dariakryvosheieva/descry-ocr

OCR for a selection of low-resource writing systems # Descry OCR Welcome to Descry, a project that brings optical character recognition technology to rare writing systems! Currently, the project supports two alphabets: **Adlam** and **Kayah Li**. More coming soon. The OCR engine uses CRAFT to detect words on an image and CRNN to recognize individual words; see the technical report for details. Model training code is available under the repo's `training` branch. Table of Contents Name App Overview Installation Guide ## Name > **descry** verb > > 1. **a:** catch sight of > **b:** find out, discover > 2. *(obsolete)* make known, reveal > > **Etymology (sense 1)**: ultimately from Latin *dēscrībere* 'describe', from *dē-* + *scrībere* 'write'. > Merriam-Webster This name was chosen because its meaning related to seeing, together with its origins in the Latin verb 'to write', make it perfect for a project that applies computer vision to writing systems. ## App Overview Use the menu on the home page to select a writing system and upload an image containing text. The OCR engine will recognize text on the image and print it in the "Model prediction" box. ## Installation Guide ### 1. Install the app on your local machine Clone the repo's `main` branch by pasting the following into your command prompt: ```shell git clone -b main --single-branch github.com ``` ### 2. Run the app To install required dependencies, open the command prompt in the **descry-ocr** folder and run ```shell pip install -r requirements.txt ``` Next, run ```shell python -m flask --app app run ``` After a few seconds, you will see `Running on 127.0.0.1`. Follow the link to open the web app in your browser.