easyocr_finetuning

This repository is a code that fine-tunes the text recognitino model of easy ocr, builds it, and uses it.

You can use it by opening easyocr_finetuning.ipynb in colab and running the cell.

1. Clone easyocr-finetuning repository

Clone the github repository.

2.Set Environment

If the problem persists even after configuring the environment, click Restart Runtime and try again.

3.Make train data

To use ai hub data, you only need to use 3-1.
If there is no ai hub data, use 3-2 to create the data.

3-1. Using AI hub OCR data

3-2. Generate Korean text image

If you do not use AI hub data, you can create your own data.
If you want to use the data, you can run the collaboratory cell and skip it if you don't want to.

4.Transform image and text to data

Only ai hub data is used here.
(If you want to use the Korean data created in 3-2, combine the folders into one.)

5.Fine tuning

Fine tuning proceeds when the cell of the collaboration is executed. You can check the learning log by looking at the saved_models folder. When learning is achieved, stop the cell and start the next cell.

6. Apply Fine tuning model to easy ocr

Import and use the fine tuning model as a custom path.

7. Evaluation

Performance is measured based on how many cropped images are fitted.